Most modern grasping systems rely on computer vision. However, computer vision is often not suitable to recover from calibration errors due to occlusion. Therefore, a recent paper on arXiv.org proposes using analytic grasp stability metrics in the task of tactile grasp refinement.
Image credit: NASA
Firstly, the hand closes its fingers in this initial grasp configuration. Then, the algorithm uses contact and finger joint position data to refine the grasp by iteratively updating the wrist and finger positions. The algorithm lifts and holds the object to evaluate the grasp’s stability.
The results show that the best results are achieved when using a quality metric based on the largest-minimum resisted wrench together with a force-based metric δ that evaluates the distance of the contact forces to the friction cone. It is also shown that that tactile sensing improves performance when training reinforcement learning agents to grasp.
Reward functions are at the heart of every reinforcement learning (RL) algorithm. In robotic grasping, rewards are often complex and manually engineered functions that do not rely on well-justified physical models from grasp analysis. This work demonstrates that analytic grasp stability metrics constitute powerful optimization objectives for RL algorithms that refine grasps on a three-fingered hand using only tactile and joint position information. We outperform a binary-reward baseline by 42.9% and find that a combination of geometric and force-agnostic grasp stability metrics yields the highest average success rates of 95.4% for cuboids, 93.1% for cylinders, and 62.3% for spheres across wrist position errors between 0 and 7 centimeters and rotational errors between 0 and 14 degrees. In a second experiment, we show that grasp refinement algorithms trained with contact feedback (contact positions, normals, and forces) perform up to 6.6% better than a baseline that receives no tactile information.
Research paper: Koenig, A., Liu, Z., Janson, L., and Howe, R., “Tactile Grasp Refinement using Deep Reinforcement Learning and Analytic Grasp Stability Metrics”, 2021. Link to the article: https://arxiv.org/abs/2109.11234
