Robotic-Arm-Based Force Control by Deep Deterministic Policy Gradient in Neurosurgical Practice

This research continues the previous work "Robotic-Arm-Based Force Control in Neurosurgical Practice". In that study, authors acquired an optimal control arm speed shape for neurological surgery which minimized a cost function that uses an adaptive scheme to determine the brain tissue force. At the end, the authors proposed the use of reinforcement learning, more specifically Deep Deterministic Policy Gradient (DDPG), to create an agent that could obtain the optimal solution through self-training. In this article, that proposal is carried out by creating an environment, agent (actor and critic), and reward function, that obtain a solution for our problem. However, we have drawn conclusions for potential future enhancements. Additionally, we analyzed the results and identified mistakes that can be improved upon in the future, such as exploring the use of varying desired distances of retraction to enhance training.