Haptic feedback rendering

Overview

Haptic feedback rendering is the step that decides what force a person feels when their hand touches something. The thing touched is a remote object in a teleoperated robot or a virtual object in a VR scene. The main use is to make grasping feel right to the person doing it.

The invariant all variants share is a loop run on every update: sense the contact, turn the sensed contact into a force, and send that force to the person's device, with limits on what the device can display. The variants differ in how contact is sensed (sensors on a robot finger, or a physics simulation), in what the force is computed from (the force alone, or the force combined with motion), and in whether any force is produced at all, since some VR methods show visual cues instead 2412.02613 §3.1 2412.02613 §3.2 2411.06244 §2.2.

Two papers with cards cover the topic. One is a survey of grasping in VR that reports no experiments 2411.06244 §1. The other is a human study that compares two rendering methods on a real glove and robot hand, with no vision 2412.02613 §4. Neither reports a shared benchmark or metric for haptic rendering quality, so their numbers cannot be compared directly 2411.06244 §2.1.

Variants

Force-only mapping

The sensed force on the robot fingertip is passed straight to the glove. Readings below a minimum threshold give no feedback, and a constant scale maps the usable sensor range onto the glove's force limit of 5 N 2412.02613 §3.2.1 2412.02613 eq. 5. The authors list this as their earlier method, and use it as the baseline. Card: Haptic Stiffness Perception Using Hand Exoskeletons in Tactile Robotic Telemanipulation.

Force scaled by motion-range factors

The force from the force-only mapping is multiplied by two factors. The first is the ratio of leader finger travel to follower finger travel during a squeeze. The second is the ratio of the maximum follower travel to the maximum leader travel, which evens out the two different motion ranges 2412.02613 §3.2.2 2412.02613 eq. 7 2412.02613 eq. 8. The authors name this variant as their proposed method, and it is the one compared against the baseline in the study 2412.02613 §5.2. Card: the same stiffness study.

Physics-based contact in VR

The grasp is produced by simulating contact and forces. The survey groups these methods as elastic-surface deformation models, haptic devices with vibration or brake feedback, and particle-based contact 2411.06244 §2.2.1. These give the most physical result, but the survey notes that checking collisions every frame is costly, and that games often favour frame rate over exact collision 2411.06244 §2.1. Card: Grasping Object: Challenges and Innovations in Robotics and Virtual Reality.

Visual stand-ins for touch

The object is snapped to the hand on contact, or the hand is frozen at the contact point. Both look stable on screen but, according to the survey, feel unnatural 2411.06244 §2.2.2. Cage-based deformation control is listed in the same group. The survey says it costs less to compute than force control 2411.06244 §2.2.2. Card: the same survey.

When to use which

The study tested only one object family, five silicone samples, and ten participants, so these rules are not tested on other hardware or tasks 2412.02613 §4.1 2412.02613 §6.

Implementation sketch

The force-only path and the motion-range path share the same first step. The sketch below follows the study's description. Units of the sensor readings are not stated, and the constants are the ones the paper gives.

# Force-only rendering  [arxiv:2412.02613 §3.2.1] [arxiv:2412.02613 eq. 1-5]
def render_force_only(fz_sensors, f_min, f_max, f_device_max):
    # Strongest z-force among the sensors on one fingertip  [eq. 1]
    contact = max(fz_sensors)
    # Readings below the minimum give no feedback  [eq. 2]
    if contact < f_min:
        return 0.0
    # Map the usable sensor range onto the device's force limit  [eq. 5]
    return contact * (f_device_max / f_max)

# Motion-range compensation on top  [arxiv:2412.02613 §3.2.2] [eq. 7-9]
def render_with_motion(fz_sensors, dz_leader, dz_follower, beta, cfg):
    base = render_force_only(fz_sensors, **cfg)
    # Leader over follower displacement; undefined before contact  [eq. 7]
    delta_z = dz_leader / dz_follower
    # Multiply the force-only output by both factors  [eq. 9]
    return base * delta_z * beta

Notes that the variants change in the paper:

• The second function needs beta eq. 8, which is the ratio of maximum travels. The study does not report the maximum values 2412.02613 §3.2.2.
• The paper does not say how dz_follower of zero is handled, and it does not cap the output at the device limit 2412.02613 §3.2.2.

For physics-based contact and visual stand-ins, the paper gives no implementation to follow. Use the survey's taxonomy as a starting point 2411.06244 §2.2.

Contradictions

• Force-only versus motion-range scaling. In the stiffness study the two methods score almost the same on the main ABX task: 74.16% and 74.58% 2412.02613 tab. 2. The subgroup results point in both directions. At distance 1 the scaled method scores 88% against 75%, but at distance 3 it scores 65% against 80% 2412.02613 §5.5. The per-group tests do not reach p < 0.05 2412.02613 §5.5. The authors conclude that the method did not give consistently significant improvements 2412.02613 §6.
• Printed binomial confidence. The study gives tail probabilities for 16 and 17 correct out of 24 that a direct binomial sum does not reproduce. The significance conclusion for 17 of 24 holds under either value 2412.02613 §5.1 2412.02613 eq. 10 2412.02613 eq. 11. Do not copy the printed percentages.

No disagreement between the survey and the study, because the survey reports no results 2411.06244 §1.

Papers

• Grasping Object: Challenges and Innovations in Robotics and Virtual Reality: survey of grasping in VR and robotics; lists rendering limits and groups VR methods by physics, visual cues and data.
• Haptic Stiffness Perception Using Hand Exoskeletons in Tactile Robotic Telemanipulation: ten-person stiffness study comparing force-only and motion-scaled haptic feedback through a glove.