HapCompass: A Rotational Haptic Device for Contact-Rich Robotic Teleoperation

1Toyota Technological Institute at Chicago (TTIC), 2University of Chicago

TL;DR: HapCompass is a wearable haptic device for contact-rich robotic teleoperation that renders 2D directional cues by mechanically rotating a single linear resonant actuator (LRA), whose asymmetric vibration provides directional haptic feedback. By mapping the robot's tactile measurements into directional feedback for the operator, it improves teleoperation performance on contact-rich tasks and leads to higher-quality demonstrations for imitation learning.

Structure and Working Principle of the HapCompass Device.

Teaser Image

HapCompass teleoperation system. (a) An operator controls the robot via hand tracking while wearing our novel HapCompass device. (b) The robot arm executes the manipulation commands. (c) We evaluate the system on three contact-rich tasks: Key Insertion, USB Insertion, and Spaghetti Probing, with (left) our method achieving higher success than (right) baseline methods.

How It Works

1. Asymmetric Vibration Creates Directional Cues

The actuator is driven by an asymmetric waveform (unlike a sinusoidal waveform), so that the vibration has a higher peak acceleration in one direction than in the opposite direction. This asymmetric vibration creates an illusory pulling sensation towards the direction of the higher peak acceleration on the fingertip.

A decoupled two-part housing design consisting of an inner (yellow) and an outer (blue) shell allows HapCompass to efficiently transmit vibrational energy to the user's fingertip, while isolating it from the device's main structure.

2. A Servo Rotates One LRA to Cover the 2D Plane

The LRA is mounted inside a rotor that is driven by a servo motor through a driving belt. By rotating the rotor to different angles, HapCompass can render directional cues across the 2D feedback plane. This design avoids relying on multiple fixed, orthogonal LRAs, as simultaneously activating several LRAs can lead to perceptual interference.

HapCompass tactile mapping overview

3. Robot Contact Forces Are Mapped Into Device Feedback

During teleoperation, the robot's tactile sensors measure changes in contact force. This delta force vector is transformed into the device frame, and then projected onto the device's 2D feedback plane. The direction and magnitude of the resulting 2D vector control the servo angle and LRA vibration amplitude, respectively.

Experiment 1: Directional Rendering Performance

We evaluated whether users can interpret directional cues rendered by HapCompass using 4-AFC and 8-AFC direction-identification tasks. Participants achieved 84.4% accuracy in the 4-AFC task and 67.2% accuracy in the 8-AFC task, both substantially above chance, showing that the device can effectively convey directional feedback overall.

Direction labels on the HapCompass device

Directional labels used in the experiment.

Directional rendering performance results

Confusion-matrix radar plots for the 4-AFC and 8-AFC direction tasks.

Experiment 2: Teleoperation Performance

We evaluated HapCompass on three contact-rich teleoperation tasks: Key Insertion, USB Insertion, and Spaghetti Probing. Across these tasks, directional tactile feedback improved success rates and generally reduced peak contact forces and bending torques compared with vision-only and non-directional tactile baselines.

Teleoperation Performance Table
Key Insertion Performance

Teleoperation performance on the Key Insertion task across four teleoperation conditions.

Demonstrations of the three contact-rich teleoperation tasks.

Experiment 3: Imitation Learning Evaluation

We conducted a preliminary imitation-learning evaluation on a simplified Key Insertion task. Policies trained on demonstrations collected with directional tactile feedback achieved higher rollout success and lower peak forces and bending torques than policies trained on demonstrations collected with non-directional controller tactile feedback, suggesting that HapCompass can improve demonstration quality for downstream learning.

Imitation learning evaluation results

BibTeX

@inproceedings{tan2026hapcompass,
  author    = {Tan, Xiangshan and Ji, Jingtian and Jiang, Tianchong and Lopes, Pedro and Walter, Matthew R.},
  title     = {HapCompass: A Rotational Haptic Device for Contact-Rich Robotic Teleoperation},
  booktitle = {Proceedings of the IEEE International Conference on Robotics and Automation (ICRA)},
  year      = {2026},
}