Imitation learning

Overview

Imitation learning fits a policy to examples of what a demonstrator or a logged system did. The core training target is the recorded action, trajectory or action token, not a reward. The Orbit paper collects teleoperated demonstrations and trains behaviour-cloning policies on them 2301.04195 §5.2 2301.04195 tab. 2. The self-driving dataset paper trains a planner to reproduce the logged future of the ego vehicle 2006.14480 §4. The shared idea is a supervised target taken from recorded actions or trajectories. The shared risk is that a small error moves the agent into states the data never covered. The driving paper names this as the accumulation of errors once the vehicle leaves the logged path 2006.14480 §2.

The same supervised idea also appears in three forms the first version of this page did not cover: a lookup over demonstration frames with no action head 2112.01511 §3.2, a transformer that receives a target return as an input token 2205.15241 §3.1, and an LLM agent that reads a trained imitator's suggested action from its prompt 2306.02224 §2.3. Imitation also applies outside robotics: a graph network learns which variable a branch-and-bound solver should split on by copying an expensive expert rule, strong branching 1906.01629 §3.3 1906.01629 §4.1.

Newer cards add three patterns. Some train diffusion policies on rollouts of a simulated expert rather than on human recordings. Some learn navigation from human targets or hand-held cameras. Some use imitation as the supervised stage of an LLM agent or a language-model-guided driver. Two survey cards mention imitation without running experiments of their own. One lists it as one of two end-to-end routes for small research cars 2404.06229 §4.4. The other groups end-to-end driving stacks under imitation, reinforcement and language-model approaches, then benchmarks three models on throughput rather than accuracy 2411.10291 §III-B 2411.10291 §V-A.

Several papers on this page add a second signal on top of the recorded data: RL fine-tuning from a reward 2207.01206 §4.3, an adversarial style reward 2601.07718 §3.5, or an inverse-dynamics model that reads actions from generated frames 2302.00111 §3.2. The reverse also occurs: an RL learner in a fixed dataset keeps its policy near the logged actions through an imitation term 2106.06860 §5. A rule-based planner can supply that recorded action during online RL, too: the parking policy REAP mixes a behaviour-cloning loss into its SAC actor update 2605.08713 §4.2. Human takeovers are a further source: HACTS records a person's corrections during a policy rollout and mixes them into the retraining data 2503.24070 §4.2. HG-DAgger goes further and lets the human decide when control switches, so the takeover itself is the gate that picks the labelled states 1810.02890 §II. Two more cards add a model-based expert or a planner as the teacher: a legged controller cloned by DAgger and then tuned with PPO 2311.02304 §III, and a navigation policy whose goal input can be masked 2310.07896 §IV-A.

Two cards sit at the edges of the page. A survey of offline RL places imitation among its method classes, as a filter on the logged data or as a policy conditioned on an outcome 2203.01387 tab. 1. It reports no experiments of its own. A broader survey of imitation learning sorts the covariate-shift fixes into methods that need an online expert, methods that keep the agent near the demonstrated states with a fixed reward, and constrained methods that avoid unfamiliar states 2309.02473 §II. Its one table lists 66 works by learning type and online-expert use 2309.02473 tab. I. Card: A Survey of Imitation Learning. A driving transformer trained on a privileged rule expert adds a rule-based speed controller on top of the learned outputs, and the authors report that removing it lowers the driving score [arxiv:2207.14024 §3.3] 2207.14024 §4.3.

Variants

Behaviour cloning on teleoperated demonstrations

Demonstrations are gathered by teleoperation with a keyboard, a gamepad or a SpaceMouse and stored in the robomimic format 2301.04195 §4 2301.04195 §5.2. The lift-cube experiment uses 2,000 trajectories for each of four settings, where the start and goal positions are either fixed or random 2301.04195 §5.2. Two policies are compared: BC and BC-RNN. The card does not describe the architecture of BC-RNN 2301.04195 tab. 2. Card: Orbit.

Author-reported success over 100 trials per setup 2301.04195 tab. 2:

Lifelong manipulation uses the same objective. LIBERO trains each task with behaviour cloning under a Gaussian-mixture negative log-likelihood, using 50 teleoperated demonstrations per task 2306.03310 §4.4. Its three policy networks (BC-RNN, BC-T and BC-ViT) all output a Gaussian-mixture action distribution 2306.03310 §4.4. Card: LIBERO.

Language-conditioned behaviour cloning across many tasks

BC-Z trains one image-to-action policy on 25,877 real-robot demonstrations covering 100 tasks. Of these, 14,769 were recorded while a learned policy ran and a person intervened 2202.02005 §4. The task is a sentence turned into a 512-dim vector by a pretrained sentence encoder, and FiLM layers inject that vector into a ResNet-18 torso 2202.02005 §5.1 [arxiv:2202.02005 §5.3]. Actions are differences from the current state to a pose N steps ahead, where N is set by an adaptive rule 2202.02005 §5.2 [arxiv:2202.02005 app. C]. Author-reported, the language-conditioned policy reaches 32% average success on held-out tasks, against 4% when the task is given as a human video 2202.02005 tab. 3. The text says 29 held-out tasks, but the table has 28 rows 2202.02005 tab. 2. The released code covers only the language route, not the video encoder Paper vs code of the BC-Z card. Card: BC-Z.

Nearest-neighbour lookup on learned features

VINN has no action network. It fine-tunes a ResNet-50 encoder with BYOL on the demonstration frames, then takes the k closest demonstration frames to a new image and averages their actions, weighted by a softmin over distance 2112.01511 §3.1 2112.01511 §3.2. The paper uses k around 10 2112.01511 §4.7. On a real door-opening task with a Hello Robot Stretch, the authors report 80% success over 30 rollouts. That compares with 53.3% for behaviour cloning on frozen representations and 0% for end-to-end behaviour cloning 2112.01511 tab. I. With the sign and the handle occluded, VINN keeps 70% against 10% for the frozen-representation baseline. With one bin also occluded, it keeps 50% against 0% 2112.01511 tab. II. Each query is compared against every demonstration frame, so the cost grows with the data 2112.01511 §4.8. Card: VINN.

Behaviour cloning on scripted or simulated experts

VIMA treats a task as a prompt that interleaves text and object images. A causal decoder predicts discrete pose actions and cross-attends to the prompt. It is trained by minimising the negative log-likelihood of expert actions over the steps of each trajectory 2210.03094 §4. The expert trajectories come from scripted oracles with privileged state 2210.03094 §3. At 200M controller parameters, the authors report 81.5% on the easiest zero-shot level against 58.0% for VIMA-Gato, and 48.6% against 16.8% on the hardest level 2210.03094 tab. 10. With 1% of the imitation data, the level-1 success falls to 36.3%; with 10% it is 76.3% 2210.03094 tab. 11. The baselines are the authors' own re-implementations 2210.03094 §5.1. Card: VIMA.

Perceiver-Actor predicts one keyframe action per step over a voxel grid built from RGB-D input, with a CLIP language goal 2209.05451 §3. Its loss is four cross-entropy terms, for translation, rotation, gripper state and collision 2209.05451 §3.4. The paper reports that the single multi-task agent beats the C2FARM-BC 3D U-Net baseline in 25 of 36 task-and-demo-count cells, with average gains of 1.33× at 10 demonstrations and 2.83× at 100 2209.05451 §4.2 2209.05451 tab. 1. The keyframe extraction is essential: the authors report zero success with random or fixed-interval keyframes 2209.05451 §4.2. Card: Perceiver-Actor.

WebShop uses imitation as the first stage of an agent. A BART model maps the instruction to a search string, trained on 1,421 instruction–search pairs 2207.01206 §4.2. A BERT choice model scores the clickable buttons, trained on 9,558 samples 2207.01206 §4.2. Reported scores on the 500-instruction test split are 59.9 with imitation alone (29.1% success) and 62.4 with imitation followed by policy gradient (28.7% success) 2207.01206 §5.2 2207.01206 tab. 2. The rule baseline reaches 45.6 and expert humans 82.1 2207.01206 tab. 2. Reinforcement learning from a BERT initialisation, without the imitation warm start, performs below the rule baseline 2207.01206 §5.2. Card: WebShop.

Diffusion-policy imitation with a pretrained visuo-tactile encoder

Touch in the Wild collects demonstrations with a handheld gripper whose soft fingers carry tactile pads, and a camera films the same moment 2507.15062 §3.1 2507.15062 §3.2. The encoder is pretrained first, by reconstructing masked tactile images from the tactile input and the CLIP embedding of the camera image 2507.15062 §4.2 [arxiv:2507.15062 eq. 1]. The policy stage then fine-tunes that encoder inside a diffusion policy that predicts noise over a 16-step action chunk, sampled with DDIM in 16 steps 2507.15062 §4.3 2507.15062 tab. 5. On four real-world tasks, with 20 trials per cell, the test-tube whole-task success is 0.85 with the pretrained encoder, 0.70 without pretraining and 0.25 for a vision-only policy 2507.15062 tab. 1. On the pencil whole task the three figures are 0.85, 0.75 and 0.45 2507.15062 tab. 1. Card: Touch in the Wild.

Co-training a diffusion policy on a large shared dataset

DROID is a dataset of 76k successful Franka demonstrations, about 350 hours, across 564 scenes, recorded by many people on one shared hardware stack 2403.12945 §1 2403.12945 §3. The policy trained on it is a diffusion policy. Two external views at 128 × 128 pass through ImageNet- pretrained ResNet-50 encoders, a frozen DistilBERT embeds the instruction, and a U-Net head denoises 16 action steps, of which 8 run open loop [arxiv:2403.12945 §5.1] 2403.12945 app. F.1. Each training batch is half in-domain demonstrations and half successful DROID episodes 2403.12945 §5.1. Author-reported, averaged over six lab, office and kitchen tasks, this co-training adds 22% absolute success in distribution and 17% out of distribution over the next-best method. Co-training with the Open X-Embodiment subset is the other comparison 2403.12945 §5.2. Per-task numbers appear only in a figure. Card: DROID.

Return-conditioned sequence models

Multi-Game Decision Transformer trains one GPT-2-style transformer with next-token prediction over sequences that hold image patches, a target return, the action and the reward 2205.15241 §3.1 2205.15241 §3.2. The training data are offline Atari trajectories from 41 games, with no filtering between expert and non-expert play 2205.15241 §3.3. At play time the model samples a high target return before each action 2205.15241 §3.4. Plain behaviour cloning is the same transformer with the return token removed 2205.15241 app. C. The authors report that the return-conditioned model beats behaviour cloning on 31 of 41 games 2205.15241 §4.7 2205.15241 fig. 8. Their headline is an interquartile mean of 126% of human-normalised score, against 101% for the training data 2205.15241 §1. The 200M-parameter model took 8 days on 64 TPUv4 chips [arxiv:2205.15241 app. B.1]. Card: Multi-Game Decision Transformers.

Imitation term in offline reinforcement learning

TD3+BC keeps a standard deterministic actor-critic algorithm and adds one imitation term to the actor loss. The actor maximises a Q term scaled by a weight λ, minus the squared error to the logged action 2106.06860 eq. 3. The weight is set per batch from α = 2.5 and the mean absolute Q value 2106.06860 eq. 5 2106.06860 §5. The paper states that these two changes are the only ones made to TD3 2106.06860 §1. Card: A Minimalist Approach to Offline RL.

On D4RL MuJoCo v0, the authors report a total normalised score of 979.3 over 15 dataset and environment pairs. That compares with 974.6 for Fisher-BRC and 764.3 for CQL 2106.06860 tab. 2. Training takes 39 minutes on one GPU against 2h 12m for Fisher-BRC and 4h 11m for CQL, using the authors' code 2106.06860 tab. 3. The gains are not uniform. On HalfCheetah-random, TD3+BC scores 10.2, below CQL (21.7) and Fisher-BRC (32.2) 2106.06860 tab. 2. The paper calls the weak random-data results a concern 2106.06860 app. C.1. The offline RL survey files this method under policy constraints, and writes the same behaviour-cloning term as its eq. 20 2203.01387 §IV-A 2203.01387 eq. 20.

Planner imitation inside an online SAC actor

REAP trains a camera-to-action parking policy with Soft Actor-Critic in a 3D Gaussian Splatting simulator 2605.08713 §4.2 2605.08713 §5.1. On each training sample, if a Reeds-Shepp planner action exists for the current state, the actor loss is a mean-squared error to that planner action with probability p. Otherwise it is the SAC actor loss 2605.08713 eq. 4 2605.08713 eq. 6. The probability p falls linearly from 1 to 0 over 8,000 epochs 2605.08713 §5.3. The critic sees privileged bird's-eye-view maps that the deployed actor does not have 2605.08713 §4.2. On standard slots the authors report 83.5% parking success, against 73.0% for SAC and 37.0% for the Reeds-Shepp curve planner 2605.08713 tab. 1. The baselines are adapted to a shared protocol and are not exact reproductions [arxiv:2605.08713 tab. 1]. In the ablation, the row labelled "w/o hybrid RL (stage 2)" drops to 0.0% success, and removing the perception pretraining gives 50.5% 2605.08713 tab. 2. The card lists the RL hyperparameters the paper does not state, including the auxiliary loss weight and the SAC temperature 2605.08713 §4.1 2605.08713 §4.2. Card: REAP.

Frozen visual features for small robot imitation heads

Voltron is a representation, not a policy. It is a masked autoencoder over video frames that either reads the paired caption as a condition or generates it, chosen per example by a single probability alpha 2302.12766 §3 2302.12766 §3.2. The encoder is then frozen, and a small head is trained on its features for each robot task 2302.12766 §4. With 25 demonstrations, the language-conditioned Voltron model reaches 38.2 ± 5.09% success on the imitation task. The variant without language reaches 33.1 ± 4.79%, and a masked-language baseline reaches 24.0 ± 4.21% 2302.12766 tab. 4. The gap to the no-language variant is about one standard deviation. Card: Voltron.

Large generalist vision-language-action policies

Octo is an open transformer policy pretrained on about 800k trajectories from Open X-Embodiment 2405.12213 §1 2405.12213 §3.2. Camera views, a language or goal-image task and readout slots become tokens, and a small diffusion head over action chunks decodes the motion 2405.12213 §3.1 2405.12213 §3.3. A new camera gets a new encoder and a new action space gets a new head, while the transformer weights are kept 2405.12213 §3.1. Each finetuning setup uses about 100 target trajectories and 50k steps 2405.12213 §3.3. Author-reported: 72% average success over six real-robot finetuning setups, against 20% for training from scratch, the best baseline 2405.12213 tab. 1. Card: Octo.

Open X-Embodiment pools 60 existing robot datasets from 34 labs and 22 embodiments into one RLDS format 2310.08864 §3.1. Each dataset's actions are normalised before they are binned 2310.08864 §4.1. RT-1-X is trained on the mixture alone, and RT-2-X is co-fine-tuned with web vision-language data in about a one-to-one split 2310.08864 §4.3. Both predict action tokens with a categorical cross-entropy loss [arxiv:2310.08864 §4.3]. The mixture weights are not stated 2310.08864 §4.3. The authors report that RT-1-X beats each dataset's original method on 4 of 5 small-data domains, but the per-dataset numbers appear only in a bar plot 2310.08864 §5.1 2310.08864 fig. 3. On the large Bridge evaluations it does not beat RT-1, while RT-2-X does 2310.08864 tab. 1. RT-2-X raises emergent-skill success from 27.3% to 75.8% against RT-2 2310.08864 tab. 2. Card: Open X-Embodiment.

OpenVLA fine-tunes a 7B vision-language model to predict discretised action tokens with next-token cross-entropy, computed only on the action tokens 2406.09246 §3.2. Its pretraining mixture contains about 970k episodes from the Open X-Embodiment collection 2406.09246 §3.3. On BridgeData V2 it reports 70.6% against 50.6% for RT-2-X 2406.09246 app. B.1.3. On two Franka tabletop tasks, full fine-tuning reaches 69.7% and LoRA of rank 32 reaches 68.2%, with 97.6M trainable parameters against 7,188.1M 2406.09246 §5.3. The two results overlap within their standard errors. The model takes one image and no proprioception, and runs at about 6 Hz on an RTX 4090 2406.09246 §3.5 2406.09246 §6. Card: OpenVLA.

FAST changes the target representation rather than the loss. Each action chunk is passed through a discrete cosine transform, scaled and rounded, then compressed with byte-pair encoding 2501.09747 §5.2. For a 50 Hz 14-dimensional chunk, the naive per-step binning needs 700 tokens and FAST needs 53 2501.09747 tab. 1. The authors report that binning makes no progress on two real-robot tasks at 20 and 50 Hz 2501.09747 §6.2. The cost is inference time: about 750 ms per chunk against about 100 ms for diffusion pi0 on an RTX 4090 2501.09747 §6.5. Card: FAST.

EventVLA adds a learned memory to a VLA. Fixed visual anchors (the first frame and recent frames) are combined with a keyframe head that predicts which future steps will be worth storing 2606.20092 §3.1 2606.20092 §3.2. On the authors' RoboTwin-MeM benchmark, anchors alone reach 18.0% and anchors with the keyframe memory reach 75.2% 2606.20092 tab. 2. On RMBench, the anchors-only variant gives the headline 67.8%, and adding the keyframe memory gives 66.7% 2606.20092 tab. 4. Inference is 0.42 s per chunk on an A100, against 1.43 s for MemER 2606.20092 tab. 11. Card: EventVLA.

DiT4DiT starts from a video generator rather than an image-text model. It pairs a video diffusion transformer with an action diffusion transformer, and trains both with flow-matching losses 2603.10448 §4.3. The action head reads hidden features from one video block at a fixed timestep, so control does not generate frames 2603.10448 §4.4 2603.10448 alg. 2. Trained from scratch on LIBERO data alone, it reports 98.6% average success, against 97.1% for OpenVLA-OFT and 96.9% for π0.5 2603.10448 tab. 1. On 24 RoboCasa-GR1 tasks it reports 50.8%, against 41.8% for GR00T-N1.5, which starts from pretrained weights 2603.10448 tab. 2. It is also the slowest of the three models in the authors' efficiency table, at 6 Hz against 9 Hz and 13 Hz 2603.10448 tab. 3. Card: DiT4DiT.

GR00T N1 is a humanoid VLA. An Eagle-2 backbone reads images and text, and a diffusion transformer trained with flow matching outputs 16-step action chunks 2503.14734 §2.1. Sampling uses four Euler steps 2503.14734 §2.1. Training mixes real, simulated, neural-generated and human-video data. Videos without motor commands get pseudo-actions from latent actions or from an inverse-dynamics model 2503.14734 §2.2 2503.14734 §2.3. The author-reported simulation average is 45.0%, against 33.4% for Diffusion Policy 2503.14734 tab. 2. On real GR-1 tasks with 10% of the post-training data, the authors report 42.6% against 10.2% for Diffusion Policy 2503.14734 tab. 3. Card: GR00T N1.

DexVLA pairs a Qwen2-VL-2B vision-language model with a roughly 1B-parameter diffusion-transformer action expert 2502.05855 §3.1 [arxiv:2502.05855 app. C.3]. Training has three stages: expert pre-training across embodiments, alignment to one embodiment, and task post-training 2502.05855 §3.2. The VLM also writes sub-step text, which the model learns to generate as intermediate output 2502.05855 §3.2. Author-reported, on their own robots: 0.92 on shirt folding with no task-specific training 2502.05855 §4.1, and 0.4 against 0.2 for pi0 on laundry folding under direct prompting [arxiv:2502.05855 §4.3]. In the expert-size ablation on shirt folding, a 93M UNet scores 0.17, a 410M diffusion expert 0.63 and a 1B expert 0.92 2502.05855 §4.4. Card: DexVLA.

Fine-tuning a VLA with parallel decoding and an L1 head

OpenVLA-OFT keeps the OpenVLA-7B pretraining and changes how it is fine-tuned 2502.19645 §3. Three changes form the recipe: all action slots of a chunk are decoded in one forward pass with bidirectional attention, a continuous head replaces the token output layer, and training minimises the mean L1 error over the chunk 2502.19645 §4.2 2502.19645 app. A.1. The LoRA rank is 32 2502.19645 tab. 4. On LIBERO, with a wrist image and proprioception on the filtered data, the authors report 97.1% average success, against 76.5% for fine-tuned OpenVLA 2502.19645 tab. 1. Throughput rises from 4.2 Hz to 108.8 Hz with parallel decoding and chunking, about 26 times 2502.19645 tab. 2. On a bimanual ALOHA robot at 25 Hz, the FiLM variant OFT+ scores at least as high as every baseline on all four tasks 2502.19645 tab. 10 2502.19645 tab. 13. On scooping it scores 100 against 93.33 for pi0, and on the pot task 51.25 against 44.17 for RDT-1B. Without FiLM, language following falls to chance (33%) on the two language-dependent tasks 2502.19645 §6.3 2502.19645 fig. 5. Card: OpenVLA-OFT.

Instruction-aware visual tokens on a frozen CLIP encoder

OTTER keeps CLIP frozen and does not fine-tune it. For each instruction word, it takes a softmax over the image patches, using the similarity between the word and each patch, and sums the patches with those weights 2503.03734 §3.1 2503.03734 eq. 4. The selected tokens are pooled per camera, and a causal transformer predicts chunks of 10-dimensional delta end-effector actions 2503.03734 §3.2. The paper states that only the pooling layers, the proprioception encoder, the transformer and the action head are trained 2503.03734 §3.2. On 70 unseen real-robot pick-and-place trials, the authors report 62% for OTTER against 12% for fine-tuned Octo and 9% for fine-tuned OpenVLA 2503.03734 §5.1. Fine-tuning CLIP itself drops unseen success to 15% 2503.03734 §5.3 2503.03734 app. C. In simulation, the unseen-task average is 61% against 48% for fine-tuned OpenVLA 2503.03734 §5.2. Card: OTTER.

Augmenting logged demonstrations to new robots

OXE-AugE does not collect new demonstrations. It cross-paints existing Open X-Embodiment frames onto other arms: the source arm is masked, the background is inpainted, and the trajectory is replayed in simulation for the target arm 2512.13100 §4.1 2512.13100 §4.2. The pooled set has 16 source datasets and over 4.4 million trajectories, with 8 or 9 target robots per dataset 2512.13100 §6 2512.13100 app. A.2. The augmented data is then used to fine-tune OpenVLA-OFT and pi0. The authors report average gains of 24% and 45% over the no-augmentation baseline on unseen robot and gripper pairs, in real tests 2512.13100 §6.2. In simulation, Diffusion Policy trained on the augmented robots is compared on five Robosuite tasks with 200 demonstrations each 2512.13100 §5. A diffusion-model version of the same augmentation lowers policy success by 27 to 30% 2512.13100 §6.2. Card: OXE-AugE.

Planning from logged or generated sequences

The driving planner is a ResNet-50 that reads a 224 × 224 bird's-eye-view raster centred on the ego vehicle and predicts its future (x, y) positions over five seconds. It uses the same squared-error loss as the forecasting baseline 2006.14480 §4. The authors follow ChauffeurNet, which augments imitation training with synthetic perturbations to limit the accumulation of errors 2006.14480 §2 2006.14480 §4. Evaluation is closed-loop on 25-second scenes. The network drives the ego vehicle, and the other agents replay their logs, so they do not react 2006.14480 §4. The paper gives no numbers for this planning result. It shows a plot and states that performance rises with the amount of training data 2006.14480 §4 2006.14480 fig. 8. Card: One Thousand and One Hours.

UniPi imitates by generating video. A text-conditioned video diffusion model produces a plan from the first observed frame and the task text, and an inverse-dynamics model reads actions from the generated frames 2302.00111 §3.1 2302.00111 §3.2. The actions are executed open-loop 2302.00111 §3.2. On seen place instructions in simulation, the authors report 59.1% against 19.4% for a state-based transformer behaviour cloner 2302.00111 tab. 1. On the multi-environment suite, the Place Bowl task gives 51.6% against 9.8% 2302.00111 tab. 3. Pretraining on internet video raises real-robot success from 72.6% to 77.1% 2302.00111 tab. 4. Training used 256 TPU-v4 chips [arxiv:2302.00111 app. A.1]. No official code was found (see the card's Paper vs code section). Card: UniPi.

Diffusion Forcing trains a causal sequence model in which every token has its own noise level, so one model can both predict and plan 2407.01392 §3.1 2407.01392 §3.2. On the single-task Maze2D benchmark, the authors report an average reward of 141.7 against 119.5 for Diffuser* 2407.01392 §4.2. In a robot task that needs memory (swapping an apple and an orange), the method reaches 80% success, and a diffusion policy without memory fails 2407.01392 §4.4. The paper warns that the reward on Maze2D can favour slow plans that stay near the goal 2407.01392 app. D. Card: Diffusion Forcing.

PGP is a motion predictor, not a control policy, but its route sampler is trained by behaviour cloning. The sampler is a learned discrete policy over a lane graph, and its loss is the negative log-likelihood of the ground-truth route 2106.15004 §4.2 2106.15004 eq. 3. The decoder that follows is trained with a winner-takes-all regression loss over sampled trajectories 2106.15004 §4.3 2106.15004 eq. 5. On nuScenes, the authors report MinADE10 of 1.00 and MissRate10,2 of 0.37, against 1.16 and 0.46 for P2T, the closest prior work 2106.15004 tab. 1. The authors report about 1 minute per pretraining epoch and 5 minutes per fine-tuning epoch on four V100 GPUs 2106.15004 app. A.7. No official code is known (PGP card). Card: PGP.

IR-WM trains its planner with an imitation term and a collision term. The imitation term measures the distance from each planned waypoint to the logged ego trajectory 2510.16729 §3.3 2510.16729 eq. 13. Its world model predicts a residual change of the bird's-eye-view state, added to the previous state, rather than the whole next scene 2510.16729 §3.2.2 2510.16729 eq. 6. At inference the default planner reads the generated BEV features, and occupancy is an auxiliary output 2510.16729 §3.2.4. On nuScenes open-loop planning, the authors report an average L2 of 0.53 m against 0.85 m for Drive-OccWorld, and a collision rate of 0.17% against 0.29% 2510.16729 tab. 4. SSR reports a lower collision rate, 0.15% 2510.16729 tab. 4. Card: IR-WM.

Closed-loop unrolled imitation for multi-agent traffic

TrafficSim learns one joint policy for every traffic actor in a scene from human driving logs. A latent variable shared by the whole scene gives several plausible joint plans from the same map and history 2101.06557 §3.2 2101.06557 eq. 1. Training runs the policy in closed loop for 12 s, with the gradient passing through the sampled actions, so errors that build up in its own rollout are penalised 2101.06557 §3.4. The imitation term is a CVAE evidence bound on the first 8 s only, and a collision penalty acts on prior samples 2101.06557 §3.4 2101.06557 app. A. The weight shifts from imitation to collision avoidance over time 2101.06557 eq. 3. On ATG4D the scenario collision rate is 0.50%, against 2.90% for the next best model, ILVM 2101.06557 tab. 1. Traffic-rule violations are 2.77% against 0.25% for IDM, so the rule metric does not favour TrafficSim 2101.06557 tab. 1. Training open-loop instead gives a collision rate of 5.92% 2101.06557 tab. 2. The authors report that a planner trained on simulated traffic reaches a collision rate of 10.73%, against 10.56% for real data 2101.06557 tab. 3. The printed time-weight formula does not give the schedule the text describes 2101.06557 app. A. No official code was found. Card: TrafficSim.

Waypoint and control heads with trajectory-guided attention

TCP is a driving network from one front camera. A shared image encoder feeds two branches. The trajectory branch predicts waypoints. The control branch predicts steering, throttle and brake for the current step and for K = 4 future steps, and at each future step it attends over the image grid with weights from the trajectory branch's hidden state 2206.08129 §3.2 2206.08129 §3.2.2 2206.08129 eq. 2 2206.08129 eq. 3. Both branches imitate Roach, a reinforcement-learning expert that sees privileged bird's-eye-view input 2206.08129 §3.1. At test time the two outputs are blended with weight 0.3 by a turning rule 2206.08129 §3.4 2206.08129 alg. 1. The authors report a driving score of 69.714 for one model and 75.137 for the ensemble, in closed-loop CARLA evaluation 2206.08129 tab. 1 2206.08129 §4.1. The repository's LICENSE is Apache-2.0, but the paper's supplement says CC BY-NC 4.0, and the code differs from the paper in several places. Check the card's Paper vs code section before reusing either. Card: TCP.

Multi-sensor transformer that imitates a privileged expert, with a safety controller

InterFuser sends three RGB views, a focus crop of the front image and a LiDAR bird's-eye-view histogram into one transformer encoder. Separate decoder queries then produce waypoints, a 20 × 20 object density map and traffic-rule outputs 2207.14024 §3.1 2207.14024 §3.2. The training data come from a rule-based expert with privileged access to the simulator, sampled at 2 FPS 2207.14024 §4.1. A safety controller reads the density map and the traffic outputs, sets a desired speed, and stops for red lights and stop signs 2207.14024 §3.3 2207.14024 app. B. The density map and traffic outputs are a readable record of what the model perceives, which the authors present as the interpretable part of the design [arxiv:2207.14024 §1].

The authors report a CARLA leaderboard driving score of 76.18, route completion of 88.23 and infraction score of 0.84, accessed June 2022 2207.14024 tab. 1. Removing the safety controller lowers the driving and infraction scores by 15% and 22%. Concatenating the views instead of fusing them with the transformer lowers them by 58% and 63% [arxiv:2207.14024 §4.3]. All evaluation is in simulation, and real-road use is left as future work 2207.14024 §5.

The released controller is a fixed rule set, not the linear programme that appendix B describes, and several loss weights and sizes differ from the paper's tables. A reimplementation that follows the paper's safety controller or loss must be written from the paper, not taken from the release. Card: InterFuser.

Style imitation alongside reinforcement learning

Hiking in the Wild trains a humanoid policy with PPO. Its style term comes from adversarial motion priors: a discriminator reads short state sequences from reference motion and the policy is rewarded for looking like it 2601.07718 §3.5. The style term has weight 0.25 and is trained with a least-squares loss 2601.07718 tab. 5 2601.07718 §3.5. Walking and running use separate reference sets, and the running reference is a single gait cycle of 1.54 s 2601.07718 app. A.9. In simulation, removing the style term drops success to 0.00% on small-box and ascending-stair terrain, against 99.09% and 100.0% with it 2601.07718 tab. 2. Card: Hiking in the Wild.

The official code is under CC BY-NC 4.0 (Hiking card), so do not reuse it commercially. This page uses only the paper's content.

Recording human corrections during a policy rollout

HACTS is a leader-follower teleoperation arm built from 3D-printed parts and low-cost servos. Its two arms are synchronised both ways, so the leader follows the robot while a policy runs, and a person can take over at any point 2503.24070 §3.1 2503.24070 §3.3. The takeovers are recorded as correction data and mixed with the original demonstrations for retraining 2503.24070 §4.2. The authors use the same arm for online RL: a pretrained actor is fine-tuned with RLPD, with short human corrections 2503.24070 §4.4. On three real UR5 tasks with ACT, the authors compare HACTS-ACT with a policy trained on 100 expert demonstrations. In the failure-correction setting, the UprightMug value is 70% against 40% for that baseline, and OpenBox and SteamBun favour HACTS-ACT in the same setting 2503.24070 tab. 2. The two arms of this comparison differ in data: HACTS-ACT gets 50 base demonstrations plus 50 correction trajectories 2503.24070 §4.2. On CloseBin, RLPD-HACTS goes from 50% to 80% after 45 minutes of online training, on 10 trials 2503.24070 tab. 4. The paper gives no official code, so the design is not reproduced here Paper vs code of the HACTS card. Card: HACTS.

Human-gated takeovers with ensemble doubt

HG-DAgger keeps DAgger's loop of training on aggregated data and rolling the novice out again. The difference is the gate: the human expert decides when to take control, instead of a coin flip with probability β, and labels are kept only while the expert is in control 1810.02890 §II [arxiv:1810.02890 eq. 2]. The novice is an ensemble. Its doubt is the l2-norm of the diagonal of the ensemble's output covariance, and the threshold is the mean of the final quarter of the doubts logged at takeovers 1810.02890 eq. 3 [arxiv:1810.02890 eq. 4]. The threshold is used only to draw a permitted set when the trained policy is evaluated. Using it as a control gate at test time is left as future work 1810.02890 §IV-B 1810.02890 §V. In simulation, the authors report fewer road departures and collisions than DAgger and behaviour cloning 1810.02890 §IV-A. On a real car with five fixed obstacle configurations, HG-DAgger has no collisions and no road departures. Behaviour cloning has one collision and six departures, and DAgger has one of each. The authors call this test too small for statistical significance 1810.02890 tab. 2 1810.02890 §IV-C. The method needs a person who can spot danger and react in real time during training [arxiv:1810.02890 §II]. Card: HG-DAgger.

Demonstrations as teaching

Pedagogical inverse reinforcement learning models a teacher who picks trajectories that make the true reward easiest for an observer to infer 2610.14004 §2.2 2610.14004 §2.3. In a goal-grid study with 59 analysed participants, people asked to show the task chose disambiguating final approaches more often than people asked to do it (1.82 against 1.05) 2610.14004 §3.2. In a feature-reward study with 58 participants, showing produced many more revisits per trajectory (1.38 against 0.10) 2610.14004 §4.2.2. Maximum-likelihood IRL recovers the reward better from showing trajectories than from doing trajectories, but the paper gives these distances only in a figure 2610.14004 §4.3. The released code has no teacher model (see the card's Paper vs code section). Card: Showing versus Doing.

Expert hints for an LLM agent

An Auto-GPT-style loop lets a chat model choose a command, call a tool, read the result and choose again. The "additional opinions" method adds a trained imitation model's proposed action to the prompt. The model still makes its own choice, and nothing is retrained 2306.02224 §2.3 2306.02224 alg. 1. The WebShop expert is a BART model that writes search queries and a BERT model that scores clickable options, both trained on demonstrations 2306.02224 §2.1.1. The ALFWorld expert is a text agent trained with DAgger 2306.02224 §2.1.2. On 50 WebShop instructions, GPT-4 with five opinions reaches 0.320 success, against 0.227 for the imitation model alone 2306.02224 tab. 1. On 134 unseen ALFWorld games, GPT-4 with the opinion reaches 0.515, against 0.306 for the imitation agent 2306.02224 tab. 2. The effect is not uniform across models. On ALFWorld, the opinion moves GPT-3.5 from 0.075 to 0.030 and Claude from 0.082 to 0.090 2306.02224 tab. 2. Card: Auto-GPT for Online Decision Making.

Imitating an expert solver's decisions

Learn2branch treats each branching decision as a classification over the fractional variables. A bipartite graph convolutional network scores every variable, and a softmax over the fractional candidates is trained with cross-entropy against the variable that strong branching chose 1906.01629 §4.1 1906.01629 §4.3 1906.01629 eq. 3. The expert labels are collected from solver runs on generated instances 1906.01629 suppl. §1. On set covering, the authors report top-1 imitation accuracy of 65.5% for the graph model, against 57.6% for the best ranking baseline, SVMrank 1906.01629 tab. 1. Solving time is lower than SCIP's default reliability pseudocost rule in 11 of the 12 problem and size settings. That count is made from the time table, not stated by the authors. On hard set covering the times are 1,489.91 s against 1,677.02 s 1906.01629 tab. 2. The models are trained on easy instances only, and the authors say they generalise to larger ones [arxiv:1906.01629 tab. 2] 1906.01629 §5.2. Card: Exact Combinatorial Optimization with Graph Convolutional Neural Networks.

Cloning a model predictive controller, then RL fine-tuning

The expert is a model predictive controller built on differential dynamic programming, with friction constraints for the stance legs and Raibert foot placement for the swing legs. It runs only in simulation, because it cannot run in real time on the robot 2311.02304 §III-A. A small MLP with layers of 512, 256 and 64 units is cloned from the expert by DAgger on flat terrain. Its input leaves out body linear velocity, because the authors found that a noisy state estimator hurt hardware 2311.02304 §III-B. A PPO stage then fine-tunes the clone on rough, step and cliff terrain, with an asymmetric actor-critic, a minimal reward and limited exploration 2311.02304 §III-C. The authors report that the fine-tuned policy survives longer than the MPC expert on rough and slippery ground, and that it tracks commands and keeps gait symmetry better than vanilla RL 2311.02304 §IV-B [arxiv:2311.02304 §IV-C] 2311.02304 §IV-D. Optimisation time falls from 0.0154 s to 0.0010 s 2311.02304 §IV-B. On the Mini Cheetah, IFM(CR) crosses a 7.5 cm step in 9 of 10 trials, and vanilla RL in 8 of 10 2311.02304 tab. 2. The hardware trial counts are small, so the authors' differences may not be meaningful 2311.02304 §IV-A. The card records no official code Paper vs code of the card. Card: Imitating and Finetuning Model Predictive Control.

Diffusion policies trained on expert rollouts

Perturbation-robust control from a simulated expert. PDP trains one RL expert per subset of motions, then rolls the experts out with noise on the executed action while storing the clean expert action as the target [arxiv:2406.00960 §3.2]. A diffusion policy trained on those pairs reaches 100.0% in-distribution and 96.3% out-of-distribution success on Bump'em perturbation recovery, against 81.0% for an MLP 2406.00960 tab. 3. The paper finds a one-step action horizon best, and a twelve-step horizon drops to 6.5% in-distribution 2406.00960 tab. 4. Inference costs K times an MLP's 2406.00960 §6. Card: PDP.

The same idea works for a wheeled robot with a planner as the expert. LiCS trains a small Transformer on the commands of the FSMT planner and of a joystick driver, with Gaussian noise on the executed commands and the clean command as the target 2406.14947 §II-A 2406.14947 §IV-A. On 66 benchmark worlds at 1.4 m/s it reaches 100% success against 99.49% for FSMT 2406.14947 tab. 2. On 16 harder worlds it reaches 91.67% against 62.50% 2406.14947 tab. 3. Its hardware run completes two of three cardboard tracks 2406.14947 tab. 5. Card: LiCS.

Legged and humanoid control can take its data from an RL expert too. An AMP-trained walking policy produces 24 datasets, varying the randomisation and the size (500K, 2M and 8M transitions) 2411.01349 §3 [arxiv:2411.01349 §5]. On flat ground the 500K sets give no stably walking policy, while the 8M "All" set matches the source policy's success of 1.0 and its tracking of 0.13 against 0.12 2411.01349 §5.1 2411.01349 tab. 3. The authors conclude that walking needs much more data than the manipulation tasks where diffusion policies are usually tested 2411.01349 §6. Card: Domain randomization for humanoid diffusion policies.

Large sets from a specialist's runs. RP1M records about one million trajectories from one DroQ specialist per piece and trains one policy on subsets of 12 to 150 songs 2408.11048 §4.3 2408.11048 §5. Diffusion Policy with a U-Net keeps in-distribution F1 between 0.45 and 0.55 across subset sizes, while the BC-MLP baseline falls from 0.529 to 0.250 2408.11048 tab. 2. On 20 unseen songs, the transformer diffusion policy rises from 0.186 to 0.316 as the subset grows 2408.11048 tab. 2. Card: RP1M.

Scaling and conditioning transformer diffusion policies

ScaleDP addresses a transformer diffusion policy that stops improving with depth. DP-T falls from 80.1% to 74.6% MetaWorld success as its layers go from 8 to 14 2409.14411 §1 2409.14411 fig. 1. ScaleDP replaces the cross-attention observation path with adaptive layer norm and removes the causal mask in action self-attention 2409.14411 §3.2. Scaled from 10M to 1B parameters, it reaches 87.50% against 51.25% for DP-T on four Franka tasks, and 98.33% against 23.33% on three bimanual UR5 tasks, with 20 trials per task 2409.14411 tab. 2 2409.14411 tab. 3. Removing the causal mask at the L size lifts the bimanual average from 66.66% to 90.00% 2409.14411 tab. 4. Card: ScaleDP.

Hierarchical and subgoal-based imitation

HDP splits the policy into two conditional diffusion models. A Guider predicts the next object contact, and an Actor denoises 16 steps of actions toward it 2411.12982 §3.2 2411.12982 tab. 8. The Actor loss adds a Q-value term weighted by η, which the authors set to 0.001 for most tasks 2411.12982 §4.4 2411.12982 tab. 9. On the real Move-T task, stage-2 success is 0.35 for Diffusion Policy and 0.65 for HDP with push data, and 0.45 against 0.75 with grasp data 2411.12982 tab. 5. The simulation baseline is a smaller Diffusion Policy that the authors re-ran, so the gains are within the paper only 2411.12982 §6.2. Card: HDP.

FeudalNav uses three levels and no RL. A latent memory map built from self-supervised features sits above a waypoint network trained on human point-clicks, and a worker turns waypoints into 15-degree turns and 0.25 m steps 2402.12498 §4.1 2402.12498 §4.2 2402.12498 §3. The human dataset has 300 trajectories and about 103,000 frames 2402.12498 tab. 1. On curved paths in Gibson, the stacked variant reaches 40.50% average success against 23.97% for NRNS without noise 2402.12498 tab. 2. Card: FeudalNav.

TaKSIE generates the next keyframe image from the current view, the instruction and a GRU embedding of past keyframes 2410.11013 §3.4.1 2410.11013 §3.3.2. Its subgoal-conditioned policy is a diffusion-based goal-conditioned behaviour-cloning policy 2410.11013 §3.4.2. Keyframes come from a progress curve over each demonstration, and regeneration happens at a cosine similarity of 0.96 or after 20 steps 2410.11013 §3.3.1 2410.11013 §3.3.3. On CALVIN the average chain length is 3.18 against 2.80 for SuSIE, and on the real robot the success rate is 71.9% against 56.3% 2410.11013 tab. 2 2410.11013 tab. 3. Card: TaKSIE.

Navigation learned from demonstrations and human targets

ViDEN collects demonstrations with a hand-held depth camera and no robot. A four-layer CNN and a four-layer transformer encode the depth image and the target, and a diffusion model over waypoints conditions on them [arxiv:2412.20226 §II-A] 2412.20226 §II-C 2412.20226 §II-D. On the hardest of three levels, it reaches 84% success, against 20% for ViNT with a relative distance goal and 12% for the BC-ConvMLP baseline, over 25 trials per level 2412.20226 tab. I 2412.20226 §III-C. Card: ViDEN.

AMR is a transformer that predicts base waypoints and camera tilt from frozen MAE tokens, LiDAR bins and goal tokens. It is trained by behaviour cloning on 500k planned trajectories in simulation, then refined with DAgger 2411.14770 §III-A 2411.14770 §III-B 2411.14770 §IV. The authors report a 3 cm and 1 degree median error on unseen objects 2411.14770 §I. A pose-estimation baseline with a CAD model and a perfect map completes 89.0% of runs when the object is visible at the start, against 95.9% for AMR 2411.14770 fig. 6 2411.14770 §V-A. Card: AMR.

LiveNet is a neural controller for two robots that imitates a receding-horizon MPC. It outputs a reference command and learned penalties for a collision barrier and a liveness barrier, and a QP layer changes only the speed 2412.04659 §4 2412.04659 §5.1. In simulation it reports no collisions and no deadlocks in 50 runs per doorway and intersection scenario 2412.04659 tab. 2. Card: LiveNet.

SAME adds a mixture of experts to the visual query of DUET's cross-attention. It is trained with DAgger on seven VLN datasets 2412.05552 §3.2 2412.05552 §4. Its R2R validation-unseen success of 76 matches the authors' multi-task ScaleVLN run 2412.05552 tab. 4. Mixing datasets in one DAgger batch drops R2R success to 62.36 from 73.69 with sequential batches 2412.05552 tab. 6. Card: SAME.

NoMaD is one Transformer diffusion policy for both exploration and goal reaching. During training the goal token is masked with probability 0.5. A masked token is removed from attention, so at test time the same network explores when the goal is hidden and heads for the goal image when it is shown 2310.07896 §IV-A. Actions are short 2D sequences sampled from a conditional diffusion model, which starts from Gaussian noise and runs K denoising steps 2310.07896 §IV-B 2310.07896 §IV-C. The paper does not state K. The authors use the policy inside a topological map of visited places, and the planner details are not in the paper [arxiv:2310.07896 §III]. Author-reported on a real robot: 98% undirected exploration success with 19M parameters, against 77% for the best published baseline with 335M 2310.07896 tab. 1. In a separate table, a dedicated Diffusion Policy also reaches 98% undirected, and goal reaching is 92% for both ViNT and NoMaD 2310.07896 tab. 2. The code is MIT-licensed, but it differs from the paper in several places, so see the card's Paper vs code section before reproducing it. Card: NoMaD.

Force-guided imitation with an impedance torque loop

FILIC feeds an estimated six-axis end-effector wrench into an ACT-style transformer policy. Two RGB streams and the wrench are fused by cross-attention, and the policy outputs target poses at 25 Hz with action chunking 2509.17053 §4.1. A joint-space impedance law runs at 2 kHz to turn each target into torque commands, so the arm yields on contact [arxiv:2509.17053 §4.1] 2509.17053 eq. 5. The wrench comes from joint torques minus the torque a digital twin predicts for the same state, multiplied by the pseudo-inverse of the transposed Jacobian, so no wrist force sensor is needed 2509.17053 §4.2 2509.17053 eq. 6. In simulated peg insertion, author-reported success is 68.0% with end-effector position only, 80.0% with joint torque and 90.0% with the estimated force 2509.17053 tab. 2. On four real tasks, the estimated-force policy reaches 76.6% to 83.3%, against 26.6% to 46.7% for position only 2509.17053 tab. 3. The released code covers only the wrench estimator README.md of the authors' repository, not the policy or the controller, and the impedance gains and training settings are not stated 2509.17053 §4.1. Card: FILIC.

Visuomotor and manipulation policies with extra inputs

RcP replaces the final pooling of a ResNet18 policy. A cross-attention query built from the joint and gripper state picks the image region that matches the robot, and a contrastive loss trains it with the L1 action loss 2411.04331 §3.1 2411.04331 §3.3. In simulated reaching under all distractors, success is 76.4% against 31.9% for Spatial-Softmax and 8.3% for R3M 2411.04331 tab. 1. On a real Franka with a self-distractor, it reaches 92% against 14% 2411.04331 §4.3. Card: RcP.

ManiWAV records sound with a contact microphone in a handheld gripper. A diffusion policy fuses CLIP vision features with audio features, and the training audio is overlaid with background sounds and robot-motor noise, each with probability 0.5 2406.19464 §3.1 2406.19464 §3.2. On wiping, the authors report 85% success against 40% for vision only, over 20 rollouts 2406.19464 §4.2. Card: ManiWAV.

Object-centric dexterous manipulation trains a transformer planner of wrist poses by behaviour cloning on human motion from ARCTIC, and a PPO finger controller adds a bounded wrist residual 2411.04005 §4.1 2411.04005 §4.2. On trained single-object trajectories in simulation, completion is 83.8% with augmentation and 59.6% without 2411.04005 tab. 4. Real-robot completion ranges from 64.3% to 100% across six objects 2411.04005 tab. 2. Card: Object-centric dexterous manipulation.

Generalist policies trained on relabelled or RL-generated rollouts

RLDG trains one RL expert per task with HIL-SERL, rolls out the successful episodes, and fine-tunes OpenVLA or Octo on them with the action log-likelihood 2412.09858 §3 2412.09858 eq. 2. The authors report gains over fine-tuning on the same number of human-demonstration episodes of 33% for OpenVLA on FMB insertion and 23% on connector insertion, and 10% and 37% for Octo 2412.09858 §4.2 2412.09858 fig. 4. On four unseen connectors, OpenVLA succeeds 73 times in 80, against 49 in 80 for the best single-connector RL policy 2412.09858 §4.3. The RL side takes one to three hours of real interaction per task and needs a success classifier 2412.09858 app. B. Card: RLDG.

Emma-X fine-tunes OpenVLA on BridgeData V2 demonstrations relabelled by a vision-language model. Each segment gets a subtask and a justification, and each frame gets a next-gripper target and a movement plan before its action tokens 2412.11974 §3.2 2412.11974 §3.3. Over 12 real WidowX tasks with 10 trials each, the authors report 57.5% success against 33.3% for OpenVLA 2412.11974 tab. 1 2412.11974 §4.2. Card: Emma-X.

Imitation for LLM agents and LLM-guided control

StepAgent pairs each expert step with the agent's own action from the same prefix. The pair becomes an implicit preference in a DPO-style loss, or it trains a discriminator whose output is a step-wise reward for PPO 2411.03817 §4.1 2411.03817 §4.2. The main results use only this step-wise reward, with no final environment reward and no human step labels 2411.03817 §7.2. On WebShop with Llama-3-8B, the inverse-RL variant scores 67.6 against 65.7 for the best baseline, ETO 2411.03817 tab. 2. Card: StepAgent.

An imitation policy with one head per route class (LEFT, MIDDLE, RIGHT) drives an RC car, and a frozen multimodal LLM answers with chain-of-thought at its own pace 2411.14256 §2.1 2411.14256 §2.2. The car keeps the last answer until a new one arrives. Trained on one cone, the policy alone succeeds 40% of the time on a harder course, against 63% to 83% with an LLM hint, depending on the model 2411.14256 tab. 3. Card: LLM-guided driving.

When to use which

The BC results come from one task in one simulator and one robot, so they do not show that BC-RNN wins on other tasks 2301.04195 tab. 2. The planner results come from one dataset collected on one route, and the closed-loop result is qualitative 2006.14480 §1 2006.14480 §4.

Implementation sketch

Common core: a supervised loss on recorded pairs, with the variants changing the data, the input and the target. Pseudocode lines are our own.

# Behaviour cloning on recorded (observation, action) pairs

def train_bc(demos, policy, optimiser, steps):
    # Demonstrations come from teleoperation
    # [arxiv:2301.04195 §5.2]
    for step in range(steps):
        obs, action = sample_pair(demos)
        # Loss and step count are not stated on the cards
        loss = action_loss(policy(obs), action)
        optimiser.step(loss)
    return policy


# Planning from logs: perturb the ego state, keep the logged target
# [arxiv:2006.14480 §4]

def perturbed_example(history, future, prob):
    # With probability prob, shift the ego state by random offsets
    if random_uniform() >= prob:
        return history, future
    offsets = gaussian_offsets()
    # The paper gives no offset magnitudes or refit method [§4]
    return history, refit_future(history, future, offsets)

Where variants differ: BC maps one observation to one action, while the planner maps a raster to a sequence of positions. VLA variants (OpenVLA, FAST) replace the action regression with a cross-entropy over action tokens 2406.09246 §3.2 2501.09747 §5.2. UniPi replaces the policy with a video model and reads actions from frames 2302.00111 §3.2. BC-RNN is the variant with the history dependence shown in the table above. VINN has no action head: it replaces the regression with a softmin-weighted average over the k nearest demonstration actions 2112.01511 §3.2. Multi-Game DT keeps the regression but adds a return token as an input 2205.15241 §3.1. TD3+BC keeps an RL actor objective and adds the squared action error to it 2106.06860 eq. 3. Learn2branch replaces the regression with a softmax cross-entropy over candidate variables 1906.01629 §4.3. Touch in the Wild replaces it with noise prediction in a diffusion policy 2507.15062 §4.3. Octo uses a diffusion head over the chunk, with 20 denoising steps 2405.12213 §3.3 2405.12213 app. D. OFT replaces the regression with a single L1 loss over the whole chunk [arxiv:2502.19645 §4.2]. GR00T N1 regresses a flow-matching velocity and integrates it in four Euler steps 2503.14734 §2.1. DexVLA's expert is a diffusion transformer whose noise schedule and loss form the paper does not state 2502.05855 §3.1. The IR-WM planner keeps a trajectory regression and adds a collision term 2510.16729 eq. 13. Voltron changes only the encoder; the task head is a small network trained on frozen features 2302.12766 §4. The additional-opinions agent keeps the LLM and adds the imitator's output as text 2306.02224 §2.3. REAP keeps a SAC actor and, for each sample, swaps in an MSE loss to the planner action with probability p 2605.08713 eq. 4.

Two design choices recur in the new cards. PDP and LiCS keep the clean expert action as the target and add noise only to the executed action, which widens the states the policy visits 2406.00960 §3.2 2406.14947 §II-A. HDP's Actor adds a weighted Q-value term to the denoising loss, with the Q-value taken on the denoised action 2411.12982 eq. 9.

Contradictions

• Pretraining. LIBERO reports that basic supervised pretraining can hurt downstream lifelong learning, though the result is shown only in a figure 2306.03310 §5.2 2306.03310 fig. 5. OpenVLA reports that its pretraining on Open X-Embodiment matters more than its fused encoder: 45.6% when trained only on BridgeData V2 against 76.3% with the full mixture 2406.09246 app. D.1. OFT's ablation points the same way for robot pretraining: 97.1 against 91.9 on LIBERO for a non-robot base 2502.19645 tab. 15. The two studies differ in model, data and metric, so their results should not be merged.

• Continuous action head. Octo reports that its diffusion head beats a continuous MSE head (83% against 35% aggregate) and a discretized head (18%) in its WidowX ablation 2405.12213 tab. 2. OFT chooses L1 regression. Its diffusion variant, run with 50 DDIM steps, appears only as an ablation, and its efficiency table reports lower success at fewer steps 2502.19645 tab. 1 2502.19645 tab. 2. The setups and the alternatives differ, so the two papers do not settle which head to use.

• pi0 on LIBERO. DexVLA's table gives pi0 an average of 97.1 2502.05855 app. A.2. OFT's table gives fine-tuned pi0 on filtered data 94.2 2502.19645 tab. 1. The DexVLA card notes that the paper does not say whether its baselines were re-run or copied, so the two figures may come from different setups.

• Per-step binning at low control rates. OpenVLA uses 256-bin per-dimension tokens at 5 Hz and 15 Hz and reports good success on its tasks 2406.09246 §5.2. FAST reports that the same binning makes no progress on tasks at 20 Hz and 50 Hz 2501.09747 §6.2. Neither paper tests the other's control rate or robot, so the findings are not directly in conflict.

• Architecture ranking. LIBERO finds that BC-T and BC-ViT beat BC-RNN on average, and that the best architecture depends on the algorithm and the suite 2306.03310 §5.2. Orbit compares only BC and BC-RNN, on a different benchmark 2301.04195 tab. 2.

• Horizon length. PDP finds a one-step horizon best for perturbation recovery and reports near-zero success at twelve steps [arxiv:2406.00960 tab. 4]. ViDEN executes 16 of 32 predicted steps and reports 84% on its hardest navigation level 2412.20226 §III-A 2412.20226 tab. I. The tasks, action spaces and controllers differ, so the two results do not settle a horizon.

• Expert versus human data. RLDG reports that successful RL-expert rollouts beat the same number of human demonstrations for OpenVLA and Octo on precise insertion 2412.09858 §4.2. LiCS's joystick-driven variant reaches 98.02% against 100% for its FSMT-expert variant in simulation [arxiv:2406.14947 tab. 4]. Both favour the non-human source on their own tasks, but the settings and margins differ, and PDP and the humanoid study report no human-data comparison 2406.00960 §5 2411.01349 §5.

• Baseline from a different codebase. HDP says the published Diffusion Policy numbers could not be matched in its setup, and it re-runs a smaller baseline 2411.12982 §6.2. Its within-paper gains do not compare with published numbers, which are not on this page.

• Diffusion-policy noise within one paper. PDP's text gives 66.9% for noisy states with noisy actions on perturbation recovery, while its table gives 59.5% 2406.00960 §5.1 2406.00960 tab. 1. Quote the table value and say so.

Pitfalls

• Compounding error. Open-loop execution is used for the video planner in all experiments 2302.00111 §3.2, and the single-image VLA takes no history 2406.09246 §6. Plan for drift when reproducing either. OFT also runs each chunk open-loop, 8 steps on LIBERO and 25 on ALOHA 2502.19645 §5.1 2502.19645 tab. 5.

• Benchmarks are not comparable. WebShop scores partial credit against a text-match reward 2207.01206 §3.1. The Maze2D reward can favour slow agents 2407.01392 app. D. LIBERO's released code picks each task's checkpoint on the same episodes it reports, so its success is optimistic (LIBERO card, Pitfalls). LIBERO averages also differ by protocol: DexVLA's 97.3 uses stage-1 weights alone [arxiv:2502.05855 app. A.2], while OFT's 97.1 fine-tunes each suite and picks the best checkpoint 2502.19645 tab. 1. Compare numbers only under the same protocol.

• Headline numbers differ within one paper. WebShop's abstract gives a 29% success rate, while its tables give 29.1% for imitation alone and 28.7% for imitation plus RL 2207.01206 §5.2 2207.01206 tab. 2. Name the model you quote.

• Small trial counts. Real-robot results use 5 to 10 episodes per task in Perceiver-Actor 2209.05451 tab. 2, 5 to 12 in OpenVLA 2406.09246 app. B.3.1, and 10 trials per terrain-gait pair in Hiking 2601.07718 §4.1. Touch in the Wild uses 20 trials per cell, so one trial moves a rate by 0.05 2507.15062 tab. 1. DexVLA uses 10 trials per task and 3 per visual-generalisation setting 2502.05855 §4.1 2502.05855 app. A.1. GR00T N1 uses 10 per real task and 5 for one task 2503.14734 §4.2. Differences of one or two trials are noise. DROID compares 10 A/B rollouts per task and method 2403.12945 §5.2.

• Best-checkpoint reporting. OFT evaluates a checkpoint every 50K steps and reports the best, on the same runs it reports 2502.19645 §5.1 2502.19645 app. A.4. GR00T N1 takes the maximum over its last five simulation checkpoints 2503.14734 §4.3. Both figures are optimistic against a fixed-step protocol.

• Averages do not follow from the columns. GR00T N1's 45.0% simulation average is not the mean of its three suites (32.1, 66.5 and 50.0 give 49.5), and its DexMG value changes between tables, 66.5% in tab. 2 and 58.5% in tab. 4 2503.14734 tab. 2 2503.14734 tab. 4. Name the table and the suite when you quote a figure.

• Flow-matching sign convention. GR00T N1's text defines the interpolation as tau times the action plus (1 − tau) times noise, with the target noise minus the action and an Euler update with a plus sign 2503.14734 §2.1 2503.14734 eq. 1. Taken literally, these do not integrate from noise to data consistently. A later release of the authors' code uses the opposite sign, which is self-consistent. Check the convention with a toy example before training (GR00T N1 card, Pitfalls).

• Wrong release in the linked repository. The repository linked from GR00T N1 holds a later release with a different backbone and an action horizon of 40, not 16 (GR00T N1 card, Paper vs code). Pin the N1 release or implement from the paper.

• Default training path may not run as released. As read, IR-WM's semi-coupled config reads a prediction key that its detector does not return, so training raises a KeyError. The paper does not mention this. The fully decoupled config uses a class that does return the key (IR-WM card, Paper vs code). Check it before reproducing the default result.

• Imitation term form differs in code. The paper writes the imitation term as a squared L2 norm 2510.16729 eq. 13. The released code takes an unsquared Euclidean distance (IR-WM card, Paper vs code). The two give different gradients, so pick one and state it.

• Released code differs from the paper. Octo's paper gives a pretraining batch of 2048, and the released config uses 512 2405.12213 §3.4. The code also clips diffusion targets and sampler iterates to [-5, 5] and scales the loss by the action dimension. The paper says nothing about either 2405.12213 §3.3 (Octo card, Paper vs code).

• Staged training needs data the release does not have. DexVLA's repository has no stage-1 training code, and its sub-step labels (Gemini 2.0 segmentation plus hand-written sub-step lists) are not released 2502.05855 app. C.2. The stage-1 learning rate and data pipeline cannot be checked against code (DexVLA card, Paper vs code).

• Expert labels depend on the states the solver visits. In learn2branch the labels come from strong branching at nodes of a tree that the paper's code grows with pseudo-cost branching, and the expert is queried at a sample of nodes 1906.01629 §4.1. The learned policy's inputs depend on that choice (learn2branch card, Paper vs code). The paper re-implements strong branching in SCIP so that it has no side effects 1906.01629 suppl. §1.

• Node counts disagree within one paper. Learn2branch's GCNN node counts for hard set covering are 29,981 in one table and 56,348 in another, while the times agree to rounding 1906.01629 tab. 2 1906.01629 tab. 3. Cite the times, and treat the node counts as uncertain.

• Seed variance. AnyTeleop's imitation results use three seeds, and several of its gaps are smaller than the standard deviations 2307.04577 tab. 3.

• Paper and code disagree in places. Examples: Perceiver-Actor's batch size (16 in the paper, 8 by default in the code), OpenVLA's data-mixture weights (the paper's shares differ from the released weights), and WebShop's test-time search choice (a random pick from the top 5 in the paper, top 1 in the code). The cards list each difference under Paper vs code (Perceiver-Actor, OpenVLA, WebShop).

• Training settings differ from the paper. Touch in the Wild's paper says the masking applies to 95% of samples independently, but the code draws it once per batch. The paper gives 60 training epochs for the policy, but the default config runs 200. In Stage 2 the code uses a CLIP learning rate of 2×10⁻⁵ where the paper says 3×10⁻⁵ 2507.15062 §4.3 [arxiv:2507.15062 §5.2]. Choose one reading and report it (Touch in the Wild card, Paper vs code).

• Released configs differ from the paper's runs. BC-Z's paper gives a multi-task learning rate of 5e-3 and batch 4096 on a TPU pod, while its released config uses 2.5e-4 and batch 32 2202.02005 app. D. DROID's default config samples with DDPM, while the paper and its generator use DDIM, and the generator leaves the 50/50 mix to the user because the two-dataset weights are commented out 2403.12945 app. F.1. The card's Paper vs code sections give the file-level evidence (BC-Z, DROID).

• Held-out count. BC-Z's text says 29 held-out tasks, while its table has 28 rows, and the card uses 28 as the count the table supports [arxiv:2202.02005 tab. 2]. Name the count when you quote its 32% figure.

• Imitation weight depends on reading. TD3+BC's λ averages |Q| over dataset actions in the paper, but over the policy's actions in the released code 2106.06860 eq. 5. The code also averages the imitation error over action dimensions, while the paper does not say whether it sums them (TD3+BC card, Paper vs code). A sum-based implementation puts more weight on imitation.

• Released token layout differs from the paper. The paper's RT-1 variant uses 8 action dimensions 2310.08864 §4.2, while the released code emits 11 tokens. RT-2-X code and checkpoints are not in that repository (Open X-Embodiment card, Paper vs code).

• Planner-dependent imitation term. REAP's anchor reward is defined by states that have a feasible planner path, and the conclusion says planner queries are still needed in training 2605.08713 §4.3 2605.08713 §6. An implementation without a planner has no anchor reward. The paper also does not say whether the planner probability is decided per episode, per step or per batch 2605.08713 §5.3.

• Training objective versus code. DiT4DiT's paper trains both diffusion transformers from one loss, but the released code stops the action gradient at the video features, so the video model learns only from its own loss. The paper also gives no value for the feature timestep or the loss weight 2603.10448 §4.3 (DiT4DiT card, Paper vs code). Choose one reading and report it.

• Two inference recipes. Multi-Game DT's main method (κ = 10, top-85 cut-off) differs from the recipe in its released code (the best of 128 return samples, top-50 cut-off) 2205.15241 §3.4 2205.15241 app. B.3. The headline 126% does not name the recipe (Multi-Game DT card, Pitfalls).

• Lookup scale is unspecified. VINN's softmin weights are exp(−distance) with no temperature, so on raw 2048-d embeddings the average can collapse onto one neighbour. The paper does not say whether embeddings are normalised 2112.01511 §3.2. Its robot results are 30 rollouts with no confidence intervals 2112.01511 tab. I.

• Partial credit in Voltron. Real-robot scores add 0.25 for each milestone reached, so they are not success rates 2302.12766 §5.4. Its control results are low in absolute terms 2302.12766 §5.3.

• Tiny test sets for the LLM agent. On 50 WebShop tasks one success is 0.02. The gap between GPT-4 with one opinion and with five (0.300 against 0.320) is one task, and the rows are single runs at low temperature 2306.02224 §3.1.1. The paper's ALFWorld text and table also swap the Claude and GPT-3.5 figures, so the table values are used here 2306.02224 §3.3.2 2306.02224 tab. 2. The expert is also trained on the same environments it advises on 2306.02224 §2.1.1 2306.02224 §2.1.2.

• Driving headline depends on a configuration. PGP's main-table row matches an encoder without GNN layers, while its decoder ablation uses two GAT layers. The card asks the reader to pick one setting and state it 2106.15004 tab. 1 2106.15004 tab. 3.

• Licences of the code. The Hiking code is CC BY-NC 4.0, so it is non-commercial (Hiking card). The DexVLA code carries a CC BY-NC style licence, so it is non-commercial too, and the card cites it for facts only (DexVLA card, Pitfalls). The pedagogical IRL repository has no licence file (Showing versus Doing card). Check the licence of each repository before reuse.

• Loss not named in OTTER. The paper does not name a training loss 2503.03734 §3.2. Its released code uses L1 unless a flag is set, but the config it reads has no such field, so the model does not construct as cloned (OTTER card, Paper vs code). Treat the loss as open until it is checked.

• Partial scores are not success rates. The OXE-AugE physical scores average 0.3, 0.5, 0.8 and 1.0 per trial 2512.13100 app. A.5, and its 24% and 45% gains have no stated unit or denominator 2512.13100 §6.2. OTTER's pick-and-place trials also give partial credit, 0.5 for the pick and 1 for the place 2503.03734 app. A.2. Compare these only with other partial-credit results.

• Augmentation assumptions. OXE-AugE assumes similar grippers and that every arm can do the task with the same strategy 2512.13100 §3 2512.13100 §7. Only "over 95%" of trajectories reach under 0.25 cm replay error for most datasets, and the rest are kept without a stated bound 2512.13100 §4.2 (2) fn. 1.

• Corrections are not a controlled comparison. HACTS attributes its gain in the failure-correction setting to correction data covering failed states, a claim the paper does not test directly 2503.24070 §4.3. Its "without any performance drop" statement rests on single 10-rollout comparisons 2503.24070 §4.3.

• Simulation-only evidence. Several headline results are simulated. PDP's results are all in simulation 2406.00960 §5. The RP1M benchmark is simulated 2408.11048 §6, as are the humanoid study [arxiv:2411.01349 §5], LiveNet's comparisons 2412.04659 §5.2 and FeudalNav 2402.12498 §8. AMR's forklift result is simulated only 2411.14770 §V-D.

• Discriminator sign in StepAgent. As written, its objective pushes the discriminator toward high scores on the agent's own pairs, and the reward is the log of that score. This is the reverse of the usual GAIL convention. The paper also uses D in three different ways 2411.03817 eq. 10 2411.03817 alg. 1 2411.03817 §5. Check the sign and the reward form before implementing the inverse-RL variant.

• Noise details are missing. PDP and LiCS add noise to executed actions and train on clean targets. PDP leaves the noise level for tracking and text-to-motion unstated 2406.00960 tab. 1, and LiCS does not give the unit of its σ = 0.25 2406.14947 §IV-A. Record the choice you make.

• Inference cost of diffusion heads. PDP's inference runs K times the cost of an MLP 2406.00960 §6. HDP runs 100 Guider and 10 Actor denoising steps per control step 2411.12982 tab. 9. ViDEN's control step also runs several U-Net passes 2412.20226 §II-D.

• Small real-robot samples. RcP's real experiment has one attempt per network and condition at each of 50 target locations 2411.04331 §4.3. TaKSIE's real-robot table uses 5 trials per task 2410.11013 tab. 3. SAME's routing and expert ablations are single runs with no seeds [arxiv:2412.05552 tab. 2]. ViDEN's 25 trials per level all come from one robot 2412.20226 §III-C.

• Baselines changed or unstated. HDP's Diffusion Policy baseline is smaller than the published one 2411.12982 §6.2. The LLM-guided driving study uses hosted models whose versions are not stated, so its numbers may not repeat 2411.14256 §3.2.

• Survey classes are not results. The offline RL survey's imitation class and its heat map collect scores from different papers, protocols and environments, and take class maxima. They show coverage, not a controlled ranking of imitation against other methods 2203.01387 §VI-C 2203.01387 fig. 5. Take the method's own card for its numbers.

Papers

• Exact Combinatorial Optimization with Graph Convolutional Neural Networks: a graph network that imitates strong branching to choose variables in a mixed-integer solver.
• One Thousand and One Hours: Self-driving Motion Prediction Dataset: an open driving dataset whose planning baseline imitates logged ego trajectories with perturbations.
• WebShop: Towards Scalable Real-World Web Interaction with Grounded Language Agents: a shopping environment whose agents are first trained by imitation, then tuned with policy gradient.
• Perceiver-Actor: A Multi-Task Transformer for Robotic Manipulation: a voxel keyframe behaviour-cloning agent for multi-task manipulation.
• VIMA: General Robot Manipulation with Multimodal Prompts: a prompt-conditioned behaviour-cloning policy trained on scripted expert data.
• Orbit: A Unified Simulation Framework for Interactive Robot Learning Environments: a simulator that collects teleoperated demonstrations and reports behaviour-cloning results on a lift-cube task.
• Learning Universal Policies via Text-Guided Video Generation: a planner that generates video from text and reads actions from the frames.
• LIBERO: Benchmarking Knowledge Transfer for Lifelong Robot Learning: a benchmark of 130 language-conditioned tasks with demonstrations, and a comparison of lifelong learning methods.
• AnyTeleop: A General Vision-Based Dexterous Robot Arm-Hand Teleoperation System: a vision-based teleoperation stack for collecting demonstrations across robots.
• OpenVLA: An Open-Source Vision-Language-Action Model: a 7B open vision-language-action model trained on robot demonstrations, with a LoRA fine-tuning recipe.
• Diffusion Forcing: Next-token Prediction Meets Full-Sequence Diffusion: a sequence model with per-token noise levels, used for planning and robot imitation.
• FAST: Efficient Action Tokenization for Vision-Language-Action Models: a compressed action tokenizer that makes autoregressive VLA training work on high-frequency dexterous tasks.
• Hiking in the Wild: A Scalable Perceptive Parkour Framework for Humanoids: a humanoid parkour policy trained with RL and an adversarial style reward.
• EventVLA: Event-Driven Visual Evidence Memory for Long-Horizon Vision-Language-Action Policies: a VLA policy with a learned keyframe memory for briefly visible cues.
• Showing versus Doing: Teaching by Demonstration: a model of teaching by demonstration, with two human studies and a reward-recovery comparison.
• The Surprising Effectiveness of Representation Learning for Visual Imitation: a nearest-neighbour lookup over demonstration frames, on a BYOL-tuned encoder.
• Multi-Game Decision Transformers: one return-conditioned transformer trained offline on 41 Atari games.
• Language-Driven Representation Learning for Robotics: Voltron, a video-caption masked autoencoder whose frozen features feed robot imitation heads.
• DiT4DiT: Jointly Modeling Video Dynamics and Actions for Generalizable Robot Control: a video diffusion transformer coupled to an action head, trained with flow matching.
• Touch in the Wild: Learning Fine-Grained Manipulation with a Portable Visuo-Tactile Gripper: a handheld tactile gripper, a visuo-tactile encoder pretrained by masked reconstruction, and a diffusion policy trained by behaviour cloning.
• Multimodal Trajectory Prediction Conditioned on Lane-Graph Traversals: PGP, a route policy trained by behaviour cloning on lane graphs, then trajectory decoding for driving.
• Auto-GPT for Online Decision Making: Benchmarks and Additional Opinions: LLM agents on WebShop and ALFWorld that receive a trained imitator's suggested action in the prompt.
• REAP: Reinforcement-Learning End-to-End Autonomous Parking with Gaussian Splatting Simulator for Real2Sim2Real Transfer: a camera-based SAC parking policy in a Gaussian-splatting simulator, with planner behaviour cloning mixed into the actor loss.
• A Minimalist Approach to Offline Reinforcement Learning: TD3+BC, a TD3 actor with a behaviour-cloning term for learning from fixed logs.
• Open X-Embodiment: Robotic Learning Datasets and RT-X Models: pooled robot datasets from 22 embodiments, and RT-1-X and RT-2-X policies trained on them.
• Octo: An Open-Source Generalist Robot Policy: a pretrained transformer policy with a diffusion action head, finetuned to new sensors and action spaces.
• DexVLA: Vision-Language Model with Plug-In Diffusion Expert for General Robot Control: a VLM with a diffusion action expert, trained in three stages with sub-step text.
• Fine-Tuning Vision-Language-Action Models: Optimizing Speed and Success: OpenVLA-OFT, a fine-tuning recipe with parallel decoding, action chunks and an L1 head.
• GR00T N1: An Open Foundation Model for Generalist Humanoid Robots: a humanoid VLA with a flow-matching diffusion transformer, pretrained on real, simulated and video-derived data.
• Vision-Centric 4D Occupancy Forecasting and Planning via Implicit Residual World Models: a driving world model that predicts residual BEV changes and plans with an imitation-plus-collision loss.
• BC-Z: Zero-Shot Task Generalization with Robotic Imitation Learning: a language- or video-conditioned behaviour-cloning policy trained on 100 real-robot tasks.
• DROID: A Large-Scale In-The-Wild Robot Manipulation Dataset: a shared-hardware manipulation dataset, and a diffusion policy co-trained with it.
• OTTER: A Vision-Language-Action Model with Text-Aware Visual Feature Extraction: a VLA that keeps CLIP frozen and selects image tokens by similarity to the instruction.
• HACTS: a Human-As-Copilot Teleoperation System for Robot Learning: a bilateral leader arm whose recorded human corrections feed imitation and RL fine-tuning.
• OXE-AugE: A Large-Scale Robot Augmentation of OXE for Scaling Cross-Embodiment Policy Learning: cross-painting of Open X-Embodiment demonstrations onto other arms, used to fine-tune OpenVLA-OFT and pi0.
• Feudal Networks for Visual Navigation: a three-level navigator trained on human point-clicks, with no RL, odometry or metric map.
• Autonomous Driving Small-Scale Cars: A Survey of Recent Development: a survey of small-scale autonomous-car platforms and driving tasks, with no experiments.
• PDP: Physics-Based Character Animation via Diffusion Policy: a diffusion policy for simulated characters, cloned from noisy rollouts of RL experts.
• LiCS: Navigation using Learned-imitation on Cluttered Space: a small Transformer that imitates a local planner for wheeled navigation, with noise-augmented data.
• ManiWAV: Learning Robot Manipulation from In-the-Wild Audio-Visual Data: a diffusion policy that fuses camera and contact-audio features, with data collected by a handheld gripper.
• RP1M: A Large-Scale Motion Dataset for Piano Playing with Bi-Manual Dexterous Robot Hands: a simulated piano dataset from RL specialists, with multi-task imitation baselines.
• Scaling Diffusion Policy in Transformer to 1 Billion Parameters for Robotic Manipulation: ScaleDP, a transformer diffusion policy with adaptive layer norm, scaled from 10M to 1B parameters.
• Incorporating Task Progress Knowledge for Subgoal Generation in Robotic Manipulation through Image Edits: TaKSIE, progress-selected visual subgoals with a subgoal-conditioned policy.
• The Role of Domain Randomization in Training Diffusion Policies for Whole-Body Humanoid Control: diffusion policies trained on humanoid walking data from an AMP expert, varying randomisation and dataset size.
• From Novice to Expert: LLM Agent Policy Optimization via Step-wise Reinforcement Learning: StepAgent, a step-wise reward from expert and agent actions for LLM agents.
• Object-Centric Dexterous Manipulation from Human Motion Data: a wrist planner trained by behaviour cloning on human motion, with a PPO finger controller.
• Raising Body Ownership in End-to-End Visuomotor Policy Learning via Robot-Centric Pooling: RcP, a pooling layer that selects the image features matching the robot's state.
• Moving Forward: A Review of Autonomous Driving Software and Hardware Systems: a survey of driving software and hardware, with a throughput benchmark of three end-to-end models.
• Hierarchical Diffusion Policy: manipulation trajectory generation via contact guidance: HDP, a two-level diffusion policy with contact guidance and a Q-learning term.
• Generalizing End-To-End Autonomous Driving In Real-World Environments Using Zero-Shot LLMs: an imitation driving policy with a frozen multimodal LLM that gives direction hints.
• Aim My Robot: Precision Local Navigation to Any Object: AMR, a transformer that drives to a centimetre-precise pose relative to an object.
• LiveNet: a neural controller that imitates MPC for two robots, with a QP safety layer.
• SAME: Learning Generic Language-Guided Visual Navigation with State-Adaptive Mixture of Experts: a mixture of experts in DUET's cross-attention, trained by DAgger on seven VLN tasks.
• RLDG: Robotic Generalist Policy Distillation via Reinforcement Learning: distils successful RL-expert rollouts into OpenVLA and Octo.
• Emma-X: An Embodied Multimodal Action Model with Grounded Chain of Thought and Look-ahead Spatial Reasoning: OpenVLA fine-tuned on demonstrations relabelled with subtasks, reasoning and movement plans.
• Embodiment-Agnostic Navigation Policy Trained with Visual Demonstrations: ViDEN, a diffusion navigation policy trained on hand-held depth-camera demonstrations.
• HG-DAgger: Interactive Imitation Learning with Human Experts: a DAgger variant where the human decides when to take over, with an ensemble doubt threshold.
• Trajectory-guided Control Prediction for End-to-end Autonomous Driving: A Simple yet Strong Baseline: a driving network with waypoint and control branches, with attention guided by the waypoints.
• NoMaD: Goal Masked Diffusion Policies for Navigation and Exploration: one diffusion policy that explores when the goal is masked and reaches a goal image when it is shown.
• Imitating and Finetuning Model Predictive Control for Robust and Symmetric Quadrupedal Locomotion: a DAgger clone of a DDP-based MPC, fine-tuned with PPO on rough terrain.
• A Survey on Offline Reinforcement Learning: Taxonomy, Review, and Open Problems: a survey that classes offline RL methods, imitation among them, and reviews benchmarks.
• Safety-Enhanced Autonomous Driving Using Interpretable Sensor Fusion Transformer: InterFuser, a multi-sensor driving transformer that imitates a privileged expert, with a rule-based speed safety controller.
• TrafficSim: Learning to Simulate Realistic Multi-Agent Behaviors: a joint traffic policy trained from driving logs by unrolling it in closed loop, with an imitation term and a collision penalty.
• A Survey of Imitation Learning: Algorithms, Recent Developments, and Challenges: a survey of behavioural cloning, IRL, adversarial IL and imitation from observation, with a table of 66 works.
• FILIC: Dual-Loop Force-Guided Imitation Learning with Impedance Torque Control for Contact-Rich Manipulation Tasks: a transformer policy with an estimated end-effector wrench and a 2 kHz impedance loop, with no force sensor.