Reinforcement learning–based tool orientation optimization for five-axis machining

Tool orientation planning is an essential process in five-axis machining for sculptured parts with complex cavity features. Unlike traditional tool orientation optimization approaches using heuristic algorithms, which demand particularly prolonged time to achieve converged optimal result, this paper presents a novel method for tool orientation optimization based on reinforcement learning algorithm. As input to the method, the rasterized feasible region is first computed via support vector machine based on an active learning fashion that requires highly reduced sampled points. The tool orientation optimization is then converted into a reinforcement learning task, in which a soft actor-critic model is utilized and trained to obtain the optimal policy. According to preliminary testing results, the proposed method is proved to be feasible for tool orientation optimization problem, and effective to produce comparable results more efficiently compared with graph-based optimization method.