DexTeleop-0: Tactile-Corrected Dual-Arm Teleoperation Contact Forces

Haichao Liu and other authors propose DexTeleop-0, a dual-arm dexterous teleoperation framework that superimposes robot-side contact corrections driven by fingertip touch on human tracking commands, to alleviate kinematic mapping errors and the problem that contact force cannot be adjusted by vision alone. The correction is computed in real time via contact stiffness approximation and the kinematic Jacobian, and adds per-contact force tracking, a soft regularizer predicting contact torque, and bounded hand joint residuals. The authors report that in controlled simulation replay, the method's final-stage success rates were 91% for Ball Assembly and 62% for Stir in Cup, versus 88% and 49% with force tracking alone. The evaluation covers simulation and real hardware, and additionally tests contact assistance using two hand types, teleoperation by seven operators, and fixed-input replay of external force trajectories.

AI 根据报道生成 · 2 days ago更新