PECHC: Robust Tactile Grasping Stabilization in Vision-Denied Peripersonal Space

Abstract

In the final “Last-Centimeter” phase of manipulation, visual occlusion and calibration errors often make vision unreliable, causing robots to suffer high failure rates due to local pose uncertainty and deviations between simulated and real-world dynamics. To address these challenges, this paper proposes the PECHC algorithm, short for Physics-Evolving Cascade Constraint and Human-Correction. To rigorously isolate the contribution of tactile feedback in multi-finger coordination, we adopt a decoupled control strategy that focuses on grasp stabilization within the hand’s workspace and functions as a fail-safe reflex. The core of the proposed approach is Hybrid Correction Imitation Learning (HCIL), which establishes a failure-triggered human–machine correction mechanism to efficiently reduce the model gap through sparse expert corrections. To improve sample efficiency and ensure reliable baseline performance, two supporting modules are introduced. Cascaded Constraint Scheduling (CCS) addresses the geometric gap by enforcing physically plausible behavioral constraints, including geometric approach, force closure, and dynamic stability. Temporal Heterogeneous Distillation (THED) addresses the physical gap by enabling implicit system identification from tactile history. Experiments demonstrate that PECHC achieves a 97.3% real-robot success rate on 150 objects from the Visual Dexterity Dataset under fully autonomous testing. In this setup, one object is used for one-time HCIL calibration, while the remaining 149 objects are evaluated without further intervention. Compared with a standard sim-to-real reinforcement learning baseline, Vanilla PPO with Domain Randomization, PECHC achieves a significant performance improvement of 42.8% and demonstrates human-like force modulation capabilities when handling fragile objects.