Integration of slip detection and grip force control in an autonomous robot assembly task for space applications

Abstract

Establishing human habitat using in-situ materials on the Moon or Mars autonomously by robotic systems is of great interest to humans. In such autonomous assembly tasks, it is crucial to integrate a slip detection capability that can detect any slipping in the grip and take corrective actions autonomously. This will enable safe handling of in-situ materials during assembly by not unnecessarily applying excessive gripping force. We construct and evaluate sliding slip detection methods (based on both heuristic and learning methods), using only the normal force recorded by a uni-axial force senor in a robotic gripper particularly from the point of view of autonomous assembly tasks for space applications. We were able to achieve over 75% sensitivity and specificity in slip detection using LSTM (Long Short-Term Memory) based model. We also observe that training the model on simulation data alone yields very robust performance in experiments and real-time prediction, and contrary to our expectation, labeling slip in the incipient phase did not yield better performance. Further, we integrate slip detection and corrective actions with an autonomous assembly task through Robot Operating System. In an experimental demonstration of autonomous assembly task using magnetic blocks, YOLO v8 for object detection and LSTM model for slip detection, our results indicate that we are able to successfully complete with correct detection of slip and corrective actions in 9 out of the 10 demonstrations.