Visual touch fusion fine operation method based on reinforcement learning

The invention discloses a visual touch fusion fine operation method based on reinforcement learning, and the method comprises the steps: processing a visual signal through a convolutional neural network, and obtaining a feature vector of visual representation; performing segmentation, feature extraction and clustering processing on the tactile sequence to obtain a feature vector of tactile representation; utilizing the joint kernel sparse coding to obtain visual touch fusion information; based on the visual contact fusion information, adopting a DDPG algorithm, training a strategy network to generate a motion track of the next step, and training a value function network to evaluate the advantages and disadvantages of the current motion track; through contact interaction with the environment, obtaining a control strategy of a specified task, and realizing optimization of an action sequence. According to the method, a robot can obtain more comprehensive external information, the information perception and fine