Learning of Efficient Stable Robot Grasping Approach Using Transformer-Based Control Policy

摘要

Measuring grasp stability is an important skill for dexterous robot manipulation task, which can be inferred from haptic information with a tactile sensor. Control policies have to detect rotational displacement and slippage from tactile feedback, and determine a re-grasp strategy in term of location and force. Classic stable grasp task only trains control policies to solve for re-grasp location with rigid object. In this work, we propose a revamped version of stable grasp task that addresses both re-grasp location and gripping force optimization aspects for objects with shifting center of gravity. We tackle this task with a model-free, end-to-end transformer-based model in reinforcement learning framework. We show that our model is able to solve both objectives after training. We also provide analysis on the performance of different models to understand the dynamic of optimizing two opposing objectives. Video results available at https://enyen.github.io/new_stable_grasp.