Skip to content
Open access

Enhancing Robotic Contact-Rich Manipulation Through Observable Passive Compliance

Oct 2026 · Machines · 0 citations · 20 references

TL;DR

Wrist deformation provides useful cues about local contact constraints, and the gains observed with larger demonstration sets suggest that these cues help policies use demonstrated contact experience more effectively.

Abstract

Small pose errors in contact-rich manipulation can cause rapid force buildup, restricting exploration and complicating demonstration collection and policy learning. This study presents observable passive compliance, which combines mechanical accommodation with deformation-based contact observation. A six-degree-of-freedom passive Stewart-platform wrist transforms brief contact transients into extended, measurable deformation, providing a “contact amplification” mechanism. Visual measurements of wrist deformation are supplied directly to an imitation-learning policy, while a learned empirical deformation envelope supports state monitoring. Force-admittance compliance (F-AC), unobserved passive compliance (Com-NoOb), and observable passive compliance (Com-Ob) are evaluated within a common learning framework using matched demonstration budgets. In a controlled probe experiment, the passive wrist reduced the maximum contact-force loading rate by 31.6%. In board and USB insertion, Com-Ob achieved significantly higher success rates than Com-NoOb with 30 and 50 demonstrations after Holm correction. With 50 demonstrations, Com-Ob succeeded in all 25 rollouts for each task, compared with success rates of 44% and 60% for Com-NoOb, respectively. Both passive configurations shortened successful demonstration and execution times relative to F-AC. An additional bolt–nut engagement experiment achieved 16 successful Com-Ob rollouts out of 25. These findings support the complementary mechanical and informational benefits of observable passive compliance. Wrist deformation provides useful cues about local contact constraints, and the gains observed with larger demonstration sets suggest that these cues help policies use demonstrated contact experience more effectively.

Read PDF

Similar papers

#artificial intelligence Preprint Oct 2026

Force-Modulated Dexterity: Unified Compliance for Contact-Rich Manipulation

Teleoperation is widely used to collect high-quality demonstrations for robot learning. Dexterous manipulation, however, requires coordinated control of motion and contact forces. Operators can specify motion, but struggle to reproduce the subtle force adjustments of human manipulation without feeling the robot's conta...

Ying Yuan, David Held, Jitendra Malik et al. · 1 citation
Preprint Oct 2026

Residual Compliant Policy: Learning Compliant Policies from Few Demonstrations

Laplacian editing is combined with simulation-based trajectory optimization to generalize across object poses, and a residual reinforcement learning policy refines the interaction forces by adjusting the reference pose of the underlying admittance controller.

Chun-Yu Jiang, Yi-Chen Cai, Shuai Li et al. · 0 citations
Preprint Oct 2026

Learning Reflexive Behavior for Contact-Rich Manipulation

During contact-rich manipulation, interactions between a robot and its environment carry information about local geometry: a surface prevents penetration, a bore guides a peg. A controller that exploits these interactions can comply with environmental constraints while preserving task intent. Robot-earning systems comm...

Quan Nguyen, Yunho Kim, Joonho Lee · 0 citations
Preprint Sep 2026

Impedance Cloning: Learning Equilibrium Point Parameters for Contact-Rich Manipulation

Contact-rich manipulation requires robots to regulate force against surfaces whose geometry deviates unpredictably from training conditions. Trajectory-based imitation learning, which reproduces observable outputs, breaks down under such shifts. We propose Impedance Cloning, which instead imitates the biomechanical pri...

Hayato Takahashi, Ryoga Oishi, Yukio Kasuga et al. · 0 citations
Preprint Sep 2026

Imp-ACT: Adaptive Impedance Control and Action Chunking with Transformers to Learn Contact-Rich Manipulation from Demonstrations

Imp-ACT is presented, a methodologically grounded and practical approach to incorporating direction-dependent Cartesian stiffness modulation directly into demonstration collection, without manual stiffness selection or offline target reconstruction.

Luca Zanetti, Doganay Sirintuna, Idil Ozdamar et al. · 0 citations
Preprint Sep 2026

Compliance for Free: Learning Identifiable Impedance via Bilateral Teleoperation

Vision-language-action models tell a robot where to move, but not how hard to push. Contact-rich tasks depend on that second quantity, compliance, yet no widely used demonstration interface records it. The obstacle is identifiability as realized pose and measured force cannot separate the operator's intended equilibriu...

H. Guda, Adrià Colomé, Carme Torras · 0 citations

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.