Dynamic switching and real-time machine learning for improved human control of assistive biomedical robots

摘要

A general problem for human-machine interaction occurs when a machine's controllable dimensions outnumber the control channels available to its human user. In this work, we examine one prominent example of this problem: amputee switching between the multiple functions of a powered artificial limb. We propose a dynamic switching approach that learns during ongoing interaction to anticipate user behaviour, thereby presenting the most effective control option for a given context or task. Switching predictions are learned in real time using temporal difference methods and reinforcement learning, and demonstrated within the context of a robotic arm and a multifunction myoelectric controller. We find that a learned, dynamic switching order is able to out-perform the best fixed (non-adaptive) switching regime on a standard prosthetic proficiency task, increasing the number of optimal switching suggestions by 23%, and decreasing the expected transition time between degrees of freedom by more than 14%. These preliminary results indicate that real-time machine learning, specifically online prediction and anticipation, may be an important tool for developing more robust and intuitive controllers for assistive biomedical robots. We expect these techniques will transfer well to near-term use by patients. Future work will describe clinical testing of this approach with a population of amputee patients.