Reference trajectory generation for force tracking impedance control by using neural network-based environment estimation

This paper presents a reference trajectory generation approach for impedance control by using neural networks to estimate the environment dynamics. In this method, the environment dynamics is estimated by a neural network (NN1), which constructs the relationship between the environment deformation and its first and second derivatives, and the interaction force. Another network (NN2) is then used to approximate the statics of the environment, which is the relationship between the interaction force and the deformation. The major advantage of the proposed method is that no exact environment model is required, so that it suites for operations on any unstructured environments. Furthermore, the neural networks have the capability of learning, due to which the precision of the generated reference trajectory will continuously be increased as the robot-environment interaction lasts. The system performance by using the proposed method is evaluated by simulations.