Abstract

Abstract—We have developed a real-time machine learning approach for the collaborative control of a prosthetic arm. Upper-limb amputees are often extremely limited in the number of inputs they can provide to their prosthetic device, typically controlling only one joint at a time with the ability to toggle their control between the different joints of their prosthesis. Many users therefore consider the control of modern prostheses to be laborious and non-intuitive. To address these difficulties, we have developed a method called Direct Predictive Collaborative Control that uses a reinforcement learning technique known as general value functions to make temporally extended predictions about a user’s behavior. These predictions are directly mapped to the control of unattended actuators to produce movement synergies. We evaluate our method during the myoelectric control of a multi-joint robot arm and show that it improves a user’s ability to perform coordinated movement tasks. Additionally, we show that this method learns directly from a user’s behavior and can be used without the need for a separate or pre-specified training environment. Our approach learns coordinated movements in real time, during a user’s ongoing, uninterrupted use of a device. While this paper is specifically focused on the control of prosthetic arms, there are many human-machine interface problems where the number of controllable functions exceeds the number of functions a user can attend to at any given moment. Our approach may therefore benefit other domains where a human and an assistive device must coordinate their efforts to achieve a goal. I.