---
kind: concept
slug: human-in-the-loop
title: Human-in-the-loop
summary: >-
  A person stays inside an AI system's decision or learning loop: choosing outputs, teaching through labels or corrections, sharing control, or limiting autonomy. Twelve cards cover these. Evidence ranges from position papers and a survey to user studies, trials and robot experiments.
tags: [human-oversight, human-ai, decision-support, agents]
papers: ["1810.02890", "1811.07871", "2105.00691", "2403.04931", "2410.03017", "2502.02649", "2503.24070", "2402.02758", "2407.14229", "2411.10016", "2411.12573", "2412.13474"]
related: [llm-agent, rlhf, active-learning, human-evaluation, prompt-engineering, human-ai-collaboration, teleoperation, imitation-learning, bayesian-optimization, finite-state-machine, model-predictive-control, impedance-control, visual-grounding, foundation-model, uncertainty-estimation]
status: draft
updated_at: 2026-10-31
---

## Overview

Human-in-the-loop (HITL) means a person is part of the cycle in which a system
decides or learns. The Hybrid Intelligence framing describes two directions.
In one, the AI supports a decision and a human evaluates its output, as when a
physician reads a model's CT analysis [arxiv:2105.00691 §3.1]. In the other,
humans teach the AI by tuning, debugging, labelling and giving feedback
[arxiv:2105.00691 §3.2]. The authors present both as parts of one arrangement
in which human and machine learn from each other [arxiv:2105.00691 §4].

The variants below differ in where the person sits: as the teacher of the
model, as the labeller of its uncertain cases, as the judge whose ratings train
a reward model, as the one who picks from the model's suggestions, as the one
who takes over a running robot policy and records corrections, as the one who
decides when a robot in training may act alone, or as the one who steps in, reviews or sets the limits on what an agent may do without
asking. Five more variants put a person in a shared-control blend, in the
labelled trials that tune a device, in the review of a robot's unsupervised
work, in the confirmation of a target before motion, and in the handover of
authority at a predicted point. The survey by Vats et al. maps the same ground
in build-time and run-time terms [arxiv:2403.04931 §2] [arxiv:2403.04931 §3.2].

## Variants

### Human as teacher of the AI

Humans label data, set or tune hyperparameters, debug models and interpret
unsupervised results such as clusters [arxiv:2105.00691 §3.2]. The paper says
interactive learning often needs many labelled examples from people, and that
human input is especially valuable when data is scarce [arxiv:2105.00691 §4].
It also names human game play as a training input for reinforcement-learning
robots [arxiv:2105.00691 §4]. The paper measures none of this. Card:
[Hybrid Intelligence](/papers/2105.00691).

### Human takes over a running robot policy and corrects it

A person watches a learned policy drive a robot, takes over when needed, and
the takeover is recorded as correction data. HACTS is a leader arm whose joints
are synchronised with the robot's: while a policy runs, the leader is driven to
the robot's joint positions, and a foot pedal switches between autonomous and
human control [arxiv:2503.24070 §3.3]. The authors name sudden movements and
lost control at takeover as problems of one-way devices, which the two-way sync
is meant to address [arxiv:2503.24070 §1]. Corrections are mixed with the original demonstrations
and used to retrain the policy [arxiv:2503.24070 §4.2].

On three real tasks with ACT, 50 base demonstrations plus 50 correction
trajectories per setting were compared with 100 expert demonstrations. On the
failure-correction setting, HACTS-ACT reached 50%, 80% and 70% on OpenBox,
SteamBun and UprightMug, against 40%, 60% and 40% for the 100-demonstration
baseline [arxiv:2503.24070 tab. 2]. The data sources are not equal in size, and
the authors attribute the gain to correction data covering failed states
[arxiv:2503.24070 §4.2] [arxiv:2503.24070 §4.3]. No controlled test isolates
that cause. Each
cell is 10 rollouts, so each result moves in 10-point steps and has no
reported variance [arxiv:2503.24070 §4.2].

For online reinforcement learning, short human corrections fine-tune a
pretrained actor with RLPD. On CloseBin, success over 10 trials rose from 50%
to 80% after 10 minutes of offline and 45 minutes of online training
[arxiv:2503.24070 tab. 4]. The paper reports no behaviour-cloning success rate
for the same task, so the gain is not compared with a baseline there
[arxiv:2503.24070 §4.5]. Card: [HACTS](/papers/2503.24070).

### Human gates control while the robot learns (interactive imitation)

Here the robot drives during training, and the human expert takes over only
when they judge the state unsafe. The expert's judgement, not a random draw,
decides who acts [arxiv:1810.02890 §II] [eq. 1]. Labels are kept only for the
states in which the expert holds control [arxiv:1810.02890 eq. 2]. DAgger, the
baseline, mixes expert and learner control with a probability β that is set
in advance. The paper says β is hard to pick beforehand and costly to tune on
a physical system [arxiv:1810.02890 §I].

The learner is an ensemble. At each takeover its doubt, the l2-norm of the
diagonal of the output covariance, is logged. The threshold τ is the mean of
the final quarter of these logged values [arxiv:1810.02890 §II] [eq. 3] [eq. 4].
The paper uses τ only to estimate a permitted set for judging the trained
policy, and it proposes but does not test using doubt as a run-time gate
[arxiv:1810.02890 §IV-B] [arxiv:1810.02890 §V] [eq. 5].

Author-reported results: in simulation, learning curves are said to be faster
and more stable than DAgger and behavioral cloning [arxiv:1810.02890 §IV-A]
[fig. 3] [fig. 4]. Inside the estimated permitted set, the collision and road
departure rates in the paper's table are about 12 and 20 times lower than
outside it, as ratios of the reported rates [arxiv:1810.02890 §IV-B]
[arxiv:1810.02890 tab. 1]. On a test vehicle with five fixed obstacle
scenarios and 5,000 samples, HG-DAgger had no collisions and no road
departures. Behavioral cloning had one collision and six departures, and
DAgger had one of each [arxiv:1810.02890 tab. 2]. The authors call the test
too limited for statistical significance [arxiv:1810.02890 §IV-C].

The method needs a human who can spot danger and react in real time during
training. The paper says it is unsuitable where that is not possible
[arxiv:1810.02890 §II footnote]. The gate is a human judgement, so nothing in
the paper automates it. The threshold formula in [eq. 4] also does not match
its own divisor as written, and the paper does not define the index base, so
the convention must be chosen by the implementer [arxiv:1810.02890 §II]. Card:
[HG-DAgger: Interactive Imitation Learning with Human Experts](/papers/1810.02890).

### Human labels the uncertain cases (active learning)

The survey describes an active-learning round in which the model predicts a
label and a confidence for each unlabelled sample. Low-confidence samples go to
human annotators, and the model is retrained on the enlarged labelled set
[arxiv:2403.04931 §2.1] [fig. 5]. The survey gives no confidence threshold and
no stopping rule for the loop [arxiv:2403.04931 §2.1] [arxiv:2403.04931 §2].
The survey runs no experiment on this loop, so its benefit is not shown here.
Card:
[A Survey on Human-AI Collaboration with Large Foundation Models](/papers/2403.04931).

### Human judges outcomes to train a reward model

The person judges what the agent achieved, and a separate reward model learns
from those judgements. An RL agent is then trained on the learned reward. The
user judges outcomes and does not have to explain how to reach them, because
credit assignment moves to the RL algorithm [arxiv:1811.07871 §3.1]. The paper
says feedback is not needed on every interaction [arxiv:1811.07871 §3.1]. It
leaves the feedback format open: preference labels, demonstrations and natural
language are all discussed, and none is picked as the default
[arxiv:1811.07871 §2] [arxiv:1811.07871 §5.5]. A reward model that is no longer
updated is a known failure, because agents learn to exploit it
[arxiv:1811.07871 §5.1]. The paper is a position paper with no experiments.
Its two assumptions, that intentions can be learned accurately and that judging
is easier than doing, are untested [arxiv:1811.07871 §1]. The authors concede
the second fails for outcomes with few possible values, such as yes/no answers,
unless an explanation is also wanted [arxiv:1811.07871 §1]. Card:
[Scalable agent alignment via reward modeling: a research direction](/papers/1811.07871).

### AI suggests, human chooses (co-pilot)

Tutor CoPilot shows a tutor one suggested reply per teaching strategy. The
replies come from GPT-4 with Bridge prompts. The tutor can edit a suggestion,
regenerate it, or pick another strategy, and the final choice stays with the
tutor [arxiv:2410.03017 §Method] [fig. 1]. Each activation sends only the 10
most recent messages to the external model [arxiv:2410.03017 §Method]. The
survey names over-trust, and missed uncertainty, as the risk of this kind of
pattern. Its suggested safeguards are showing confidence, explaining outputs
and allowing override [arxiv:2403.04931 tab. 2]. The survey does not test them.

In a randomized trial the tutors were randomized, not the students, with 783
tutors analysed [arxiv:2410.03017 §Methods]. Access to the tool raised the
unconditional exit-ticket pass rate by 4.0 percentage points, from 61.7% to
65.7% [arxiv:2410.03017 tab. 1]. The authors report a larger 14-point figure
from an instrumental-variable estimate. That figure applies to tutors whose use
changed with assignment, not to the tool overall [arxiv:2410.03017 SI: Treatment
on the Treated Analysis]. The gain was larger for students of lower-rated tutors
[arxiv:2410.03017 §Results]. The authors put the cost at about USD 20 per tutor
per year [arxiv:2410.03017 SI: Study Costs]. Card:
[Tutor CoPilot](/papers/2410.03017).

### Human sets the autonomy limit (agents)

This variant is a position, not a method. The authors propose a five-level
scale of agent autonomy, from a program in which the model has no say over the
flow to one in which the model creates and runs new code [arxiv:2502.02649
tab. 1]. They argue that fully autonomous agents should not be developed
[arxiv:2502.02649 §1] [arxiv:2502.02649 §6]. They judge semi-autonomous systems
to balance risks and benefits better, but name no threshold for where the line
lies [arxiv:2502.02649 §1]. Their recommendations are a spectrum of autonomy
levels, human-control mechanisms, and methods that keep agents within
constraints set by humans [arxiv:2502.02649 §6]. Card:
[Fully Autonomous AI Agents Should Not be Developed](/papers/2502.02649).

The survey adds a run-time view of the same question. It names five
collaboration patterns, among them dynamic delegation with a control handoff
and supervisory orchestration with plan review and approval checkpoints. For
each it lists the risk and a suggested safeguard, such as escalation rules,
readiness checks and audit logs [arxiv:2403.04931 tab. 2]. The risks it names
are a handoff that comes too late or too often, and errors that build up
unseen in multi-step tasks [arxiv:2403.04931 tab. 2]. The survey proposes no
thresholds for these, and it reports no tests of the safeguards
[arxiv:2403.04931 §3.2]. Card:
[A Survey on Human-AI Collaboration with Large Foundation Models](/papers/2403.04931).

### Human shares control of a robot (blended command)

A shared-control arm blends the human's command with the robot's own
reference, and one weight sets how much the robot decides. The commanded
position is a convex combination with weight gamma on the robot
[arxiv:2402.02758 §3.2] [arxiv:2402.02758 eq. 3]. In a 24-person study of
teleoperated 3-D tracking, the statistics compare a low level (gamma 0.2) with
a high level (gamma 0.8) [arxiv:2402.02758 §4].

Self-reported workload (NASA-TLX) fell and self-reported trust (MDMT) rose at
the higher level, both with p < .001 [arxiv:2402.02758 tab. 1]. The two
objective load measures, tapping error and pupil index, showed no autonomy
effect [arxiv:2402.02758 tab. 1]. So the self-report and the objective
measures disagree in this study. Tracking error was lower at the high level,
0.83 cm against 2.16 cm in the units the paper gives [arxiv:2402.02758 tab. 2].
The high level is gamma 0.8, not full autonomy, which is gamma 1
[arxiv:2402.02758 §3.2]. The paper notes that the robot gave no haptic feedback
and that the design covers one task [arxiv:2402.02758 §5.4]. Card:
[Effects of Shared Control on Cognitive Load and Trust in Teleoperated Trajectory Tracking](/papers/2402.02758).

### Human's labelled trials tune a device

A threshold-based state machine detects walk-to-sit and stair transitions on
two lower-limb exoskeletons, and its thresholds were first set offline
[arxiv:2411.12573 §2.1] [arxiv:2411.12573 §2.2]. For a user whose transitions
are missed, the thresholds can be re-set with Bayesian optimization (BO)
against that user's labelled transitions [arxiv:2411.12573 §2.5.2]
[arxiv:2411.12573 alg. 1]. The BO cost is computed from ground-truth labels, so
the user must supply labelled data before deployment [arxiv:2411.12573 alg. 2].
A statistics-based rescaling (SBA) also adapts the thresholds without a search
[arxiv:2411.12573 §2.5.1].

On eWalk, BO raised the W-SD and SD-W accuracies to 96% and 98%
[arxiv:2411.12573 §4.2] [fig. 6]. On autonomyo, BO raised four transitions from
77.5, 75, 77.5 and 75% to 90, 90, 87.5 and 82.5% [arxiv:2411.12573 §4.2]
[fig. 6]. SBA raised three eWalk transitions from 92, 86 and 80% to 94, 90 and
88%, and gave no improvement on autonomyo [arxiv:2411.12573 §4.2]. One
subject's W-SD transition on eWalk went from 20% to 100% [arxiv:2411.12573
§4.2] [fig. 7]. The abstract's "up to 80%" figure is not tied to a table, and
this single subject is the closest in-text case [arxiv:2411.12573 §4.2].

BO was not run in real time [arxiv:2411.12573 §5.5]. The paper estimates about
2 minutes on a PC and does not measure an embedded run [arxiv:2411.12573 §5.4].
Ground truth was labelled live by an observer during zero-torque trials, and
the tests gave no active assistance [arxiv:2411.12573 §3] [arxiv:2411.12573 §5.5].
Card: [Personalising exoskeleton transition thresholds](/papers/2411.12573).

### Human reviews a summary of unsupervised robot work

When a robot works without supervision, its operator must learn afterwards
what it did. Pretrained video models can summarise a 40-minute robot video as
a generic summary, chosen with no input, or as a query-driven summary, chosen
after the operator asks a question. Each summary can be shown as a storyboard,
a short video or text [arxiv:2411.10016 §3]. No model was fine-tuned
[arxiv:2411.10016 §3.4.3].

In a 30-person study with four 40-minute videos, query-driven summaries raised
answer accuracy over raw video (coefficient 0.549, p = 0.005) and added task
time (156.3, p = 0.002) [arxiv:2411.10016 tab. 2]. Generic summaries did not
significantly change accuracy [arxiv:2411.10016 tab. 2]. Among modalities,
storyboards were more accurate than video (0.905, p = 0.004) and took less time
(-124.7, p = 0.028). Text was less accurate than video (-0.955, p = 0.005).
The storyboard advantage did not hold for event questions (interaction
-2.447, p < .001) [arxiv:2411.10016 tab. 2]. Time units are not stated
[arxiv:2411.10016 tab. 2].

The four videos were paired with tasks, but which video went with which
modality is not stated, so video and modality effects may be confounded
[arxiv:2411.10016 §3.3]. Per-condition accuracy percentages are not reported
[arxiv:2411.10016 §4.4]. Card: [Video foundation models for intermittent robot supervision](/papers/2411.10016).

### Human confirms a target before the robot moves

An operator gives a spoken or written instruction, such as where a hand should
rest, and the system shows a candidate point on the robot's camera image. Each
later instruction is treated as a prediction, a correction or a confirmation,
and only the confirmed point goes to the controller [arxiv:2407.14229 §3]
[arxiv:2407.14229 §3.2] [fig. 2]. A language model writes the arithmetic that
places the point relative to detected objects, and a segmenter handles named
objects [arxiv:2407.14229 §3.3].

On 78 image, prompt and mask tuples, the best combination (GPT-3.5 with
GroundingDINO and CLIPSeg) reached a median overall success of 0.61, with
quartiles of 0.51 and 0.70, against 0.13 for random points [arxiv:2407.14229
tab. 1]. Cells use different models and the paper does not state the sample
size per cell [arxiv:2407.14229 §4.1]. The masks mark acceptable contact areas,
but the rule for a hit is not stated [arxiv:2407.14229 §4.1] [fig. 7]. In a
pilot with 11 volunteers, the median distance to the target at the fifth prompt
was 21 px, against an 18 px target radius [arxiv:2407.14229 §4.2] [fig. 8].
On the Talos humanoid, all four reaching settings succeeded, with no trial
counts given [arxiv:2407.14229 §4.3]. The released code holds only the
language and vision part, so the interaction loop must be built from the paper;
see the card's Paper vs code section. Card:
[Words2Contact: Identifying Support Contacts from Verbal Instructions Using Foundation Models](/papers/2407.14229).

### Human hands authority over at a predicted point

A planner models human reaching as a trade-off between speed and accuracy,
and between effort and reward, then solves for a smooth motion
[arxiv:2412.13474 §2.1] [arxiv:2412.13474 §3.3]. In co-manipulation, the robot
estimates the human's goal from the first 0.2 s of motion and then runs the
planned trajectory open-loop [arxiv:2412.13474 §5.1]. The solve takes 1.3 s for
the seven-DoF arm [arxiv:2412.13474 tab. 1].

For a task with a fast transport and a slow precise part, a Gaussian process
fitted to earlier human data predicts where a person switches between them
from the distance to the goal [arxiv:2412.13474 §2.2] [fig. 2]. At that point
the robot's stiffness drops from 400 N/m to 0, handing control over
[arxiv:2412.13474 §5.2].

The evidence is qualitative. The handover test used one operator and five
trials per goal, and no baseline controller was run [arxiv:2412.13474 §5.1]
[arxiv:2412.13474 §5.2]. In simulation, planned velocities follow width and
distance trends by eye, with no fit reported [fig. 3]
[arxiv:2412.13474 §3.4]. Card: [Planning Human-Robot Co-manipulation with Human Motor Control Objectives and Multi-component Reaching Strategies](/papers/2412.13474).

## When to use which

| Situation | Use | Why |
|---|---|---|
| Little labelled data, and a model must be trained or tuned | Human as teacher | The paper says human input matters most when data is scarce, but reports no measurement [arxiv:2105.00691 §4] |
| A person must decide on each output, and an expert-like suggestion would help | AI suggests, human chooses | One randomized trial found a 4.0-point gain on exit tickets at about USD 20 per tutor per year; the effect varied by tutor [arxiv:2410.03017 tab. 1] [arxiv:2410.03017 SI: Study Costs] |
| An agent can act with side effects that may reach people | Cap autonomy and keep a human in control | The authors argue risk to people grows with autonomy, and give no threshold [arxiv:2502.02649 §1] [arxiv:2502.02649 §5.2] |
| The task has no written reward, and a person can judge outcomes more easily than specify behaviour | Human judges outcomes to train a reward model | Untested position; the authors concede the judging assumption fails for yes/no-style outcomes [arxiv:1811.07871 §1] |
| A learned robot policy fails on states it did not see in training, and a person can take over on the same robot | Human takes over and corrections join the training data | Author-reported, 10 rollouts per cell: FCID success rose to 50%, 80% and 70% on three tasks against 40%, 60% and 40% for 100 expert demonstrations [arxiv:2503.24070 tab. 2] |
| A robot learns by imitation, and an expert can take over safely in real time during training | Human gates control and labels only the takeovers (HG-DAgger) | Author-reported, one test vehicle, five scenarios: 0 collisions and 0 road departures, against 1 and 6 for behavioral cloning and 1 and 1 for DAgger [arxiv:1810.02890 tab. 2] |
| A model's uncertain predictions need labels, and annotation is the bottleneck | Human labels the uncertain cases | The survey describes this loop but gives no threshold and no test of its benefit [arxiv:2403.04931 §2.1] |
| An agent runs multi-step tasks, and a human must be able to stop or take over | Escalation and approval checkpoints | The survey names the risks and suggests safeguards, without thresholds or tests [arxiv:2403.04931 tab. 2] [arxiv:2403.04931 §3.2] |
| The outcome has no clear ground truth | Not covered | The authors say such management problems are time-variant and hard to score, and propose no alternative [arxiv:2105.00691 §6] |
| Fully autonomous operation with no human check | Not covered | The position paper argues against building it [arxiv:2502.02649 §6] |
| A robot shares a motion with a person, and the person's workload and trust matter | Shared control with a blend weight | Self-reported load and trust moved with the robot's share in 24 people; two objective load measures did not [arxiv:2402.02758 tab. 1] |
| A device must fit one user's gait, and that user can give labelled transitions first | Bayesian optimization on the user's labels | BO raised accuracy on both exoskeletons; the rescaling helped on one only; BO needs labels and is not real-time [arxiv:2411.12573 §4.2] [arxiv:2411.12573 §5.5] |
| An operator must review a long run they did not watch | Storyboard summary, or a query-driven one | Storyboards were the most accurate modality but weaker on event questions; query-driven summaries raised accuracy at a time cost [arxiv:2411.10016 tab. 2] |
| A robot target comes from language, and a wrong contact is costly | Operator confirms the candidate point | Best combination reached a median of 0.61 on 78 samples, against 0.13 for random points; the hit rule is not stated [arxiv:2407.14229 tab. 1] [arxiv:2407.14229 §4.1] |
| Two bodies move together, and control must pass to the person at a predictable point | Handover at a Gaussian-process-predicted point | Qualitative only: one operator, five trials per goal, no baseline [arxiv:2412.13474 §5.1] [arxiv:2412.13474 §5.2] |

## Implementation sketch

The papers give no interfaces, so the sketch below is our reading of the
co-pilot loop plus a feedback log. The names `propose`, `human.decide` and
`log` are placeholders.

```pseudocode
# Human-in-the-loop step: the AI proposes, the human decides, feedback is kept
# [arxiv:2410.03017 §Method] [arxiv:2105.00691 §3.2]
# Input:  context, options (e.g. strategies), human, log
# Output: the final action, which the human chose or edited

def human_in_the_loop_step(context, options, human, log):
    # One candidate per option, built from the live context  [arxiv:2410.03017 §Method]
    candidates = {opt: propose(context, opt) for opt in options}

    # The human edits, regenerates or picks another option; their choice wins
    # [arxiv:2410.03017 §Method] [fig. 1]
    final = human.decide(candidates)

    # Keep the context, the candidates and the choice as labelled data
    # [arxiv:2105.00691 §3.2]
    log.append(context, candidates, final)
    return final
```

## Contradictions

No direct numerical conflict. The papers mostly test different things, and
the points of tension are these:

- **Trust and autonomy.** The shared-control study finds self-reported trust
  rises as the robot's share grows [arxiv:2402.02758 §4.3]. The position paper
  argues that fully autonomous agents should not be developed [arxiv:2502.02649
  §6]. The study's high level is not full autonomy, and its robot is an arm,
  not an agent [arxiv:2402.02758 §3.2]. The two do not test the same claim.
- **Load measures.** In the shared-control study, self-reported workload fell
  with autonomy while tapping error and pupil index did not [arxiv:2402.02758
  tab. 1]. The study does not say which measure is the better guide to load.

The Hybrid Intelligence claim that combined results are superior to either part
[arxiv:2105.00691 §4] is not tested. Tutor CoPilot measures one gain in one
setting [arxiv:2410.03017 tab. 1]. Do not read one as confirming the other.

## Papers

- [Scalable agent alignment via reward modeling: a research direction](/papers/1811.07871): position paper proposing that a user's feedback trains a reward model that an RL agent then optimises; no experiments.
- [Hybrid Intelligence](/papers/2105.00691): position paper defining humans and AI as complementary partners that learn from each other; no experiments.
- [A Survey on Human-AI Collaboration with Large Foundation Models](/papers/2403.04931): survey of build-time guidance and run-time collaboration patterns for foundation models; no experiments.
- [Tutor CoPilot](/papers/2410.03017): a tutor-facing assistant that suggests one reply per teaching strategy; a randomized trial with 783 tutors, author-reported.
- [Fully Autonomous AI Agents Should Not be Developed](/papers/2502.02649): position paper proposing a five-level autonomy scale and arguing against full autonomy; no experiments.
- [HACTS: a Human-As-Copilot Teleoperation System for Robot Learning](/papers/2503.24070): a leader-follower arm that lets a person take over a running policy, with corrections used for imitation learning and RLPD; author-reported results on a UR5 with 10 rollouts per cell.
- [HG-DAgger: Interactive Imitation Learning with Human Experts](/papers/1810.02890): a DAgger variant in which the human expert decides when to take control and labels only those takeovers, with a doubt threshold from ensemble variance; author-reported simulation and five-scenario vehicle tests.
- [Effects of Shared Control on Cognitive Load and Trust in Teleoperated Trajectory Tracking](/papers/2402.02758): a 24-person study of a blended-command robot arm; self-reported load and trust moved with autonomy, objective measures did not; author-reported.
- [Personalising exoskeleton transition thresholds](/papers/2411.12573): a threshold state machine for two exoskeletons, personalised by statistics or by Bayesian optimization on labelled user data; author-reported.
- [Video foundation models for intermittent robot supervision](/papers/2411.10016): pretrained video models summarise 40-minute robot runs for a supervisor; a 30-person study, author-reported.
- [Words2Contact: Identifying Support Contacts from Verbal Instructions Using Foundation Models](/papers/2407.14229): an operator's instructions and corrections pick a contact point on the camera image before the robot moves; author-reported on 78 samples.
- [Planning Human-Robot Co-manipulation with Human Motor Control Objectives and Multi-component Reaching Strategies](/papers/2412.13474): a planner built on human motor-control objectives, with goal estimation and a predicted handover point; qualitative, one operator.