Robotic Execution of Contact-Rich Manipulation via Teleoperation or Imitation Learning

Determine effective approaches that enable robots to perform contact-rich manipulation tasks through either direct teleoperation or policies learned from human demonstrations (imitation learning), ensuring reliable execution in tight-clearance insertions and other contact-intensive scenarios.

Background

The paper motivates the difficulty of contact-rich manipulation, such as tight-clearance insertions and low-visibility assembly, noting that current teleoperation interfaces often lack informative directional haptic feedback and that this gap can also degrade the quality of demonstration data for imitation learning.

HapCompass is proposed as a step toward addressing this challenge by providing 2D directional haptic cues to operators. While the device and system improve teleoperation performance and demonstration quality in the reported studies, the broader challenge of reliably enabling robots to perform contact-rich tasks via teleoperation or imitation learning remains a stated open problem.

References

However, enabling robots to perform these tasks, either through direct teleoperation or by learning from human demonstrations (i.e., imitation learning), remains an open problem.

— HapCompass: A Rotational Haptic Device for Contact-Rich Robotic Teleoperation  (2603.30042 - Tan et al., 31 Mar 2026) in Section 1, Introduction

LEVER remains the hardest at 50\% because both commands are contact-critical presses and either failure invalidates the trial.

— Facet-0: A Robotic Foundation Model for Contact-Rich Precise Manipulation  (2609.01596 - Deng et al., 1 Sep 2026) in Section 4.1, “Main Comparison and Component Gains (Q1)”