• optical flow
• tactile flow
• optical tactile sensor
• sensory motor control
• robotic manipulation

The research delves into leveraging optical flow to extract tactile flow information and develop a deep learning-based estimator (neural network), with the goal of accurately replicating the human tactile sense. The study encompasses dataset creation, neural network training, and real-time application using the DIGIT sensor on the robot to mimic the sense of touch found in humans. To emulate the entire somatosensory system on the robot, a Kalman filter is deployed to effectively balance the contributions of proprioception and exteroception (tactile flow obtained from the neural network). Through simulation of human behavior using the neural network and Kalman Filter, the system mirrors human perception in reaching specific points on ridged surfaces. Acknowledging biases in human touch induced by ridges, an MPC is implemented to assess the optimal ridge orientation, allowing the robot to reach a desired point while perceiving itself as sliding towards a virtual point