Human-Inspired Haptic-Based Planning and Control for Robotic Sliding Movements

This work presents a human-inspired haptic-based framework for planning and controlling a robotic manipulator during sliding movements over ridged surfaces. We use the DIGIT soft optical tactile sensor to detect surface textures, and deep learning techniques to approximate tactile flow by evaluating optical flow (OF) from collected images. A dataset is created to train a neural network (RAFT) to compute dense OF fields, representing the tactile perception component. The robot motion control mimics human unbiased musculoskeletal proprioception, using Model Predictive Control (MPC) and a Kalman Filter to adjust trajectory based on tactile feedback. Experimental results show promising performance in real-time control. The framework demonstrates potential for future applications in robotic manipulation and tactile feedback systems.