Line-grasping control of de-icing robot based on k-nearest neighbor reinforcement learning
基于k–最近邻分类增强学习的除冰机器人抓线控制

The flexible mechanical characteristic of power lines induces difficulties for line-grasping control for de-icing robots. To deal with this difficulty, we propose for de-icing robots a line-grasping control approach which combines the k-nearest neighbor (KNN) algorithm and the reinforcement-learning (RL). In the learning iteration, the state-perception mechanism of the KNN algorithm selects k-nearest states and weights; from k-weighted states, an optimal action is determined. By expressing a continuous state by k-nearest discrete states in this way, this approach effectively ensures the convergence for the computation and avoids the curse of dimensionality occurred in traditional continuous state-space generalization methods. Abilities of RL in perception and adaptation to the environment make the line-grasping control to tolerate possible errors in robot model, errors of robot arm attitudes and interferences from the environment. The design procedures are presented in details. Simulation results of line-grasping control based on this approach are given.