Toolpath Calculation Using Reinforcement Learning in Machining

Artificial intelligence is now used in many domains. Reinforcement learning is one of three basic machine learning paradigms, alongside supervised learning and unsupervised learning. Calculating toolpath in machining can be fastidious and often pass through a selection of an extensive set of options and parameters. Automatic toolpath calculation has been a challenge for decades. Reinforcement learning is a good candidate to teach an agent “good trajectories” and then let the agent determines the trajectory for any new case. In the proposed paper, reinforcement learning has been tested to learn toolpath calculations in turning. In a first attempt, simple tuning part has been tested. The paper will present the environment, the different codification, the reinforcement learning algorithm and the reward systems proposed and tested. It has been verified, for all these options, that the agent learns and then that it can reapply what it learns on new parts. The paper concludes that reinforcement learning is a promising direction for toolpath calculation and draw some perspective toward its generalisation in the CAM systems.