Effective Neural Representations for Brain-Mediated Human-Robot Interactions

Abstract

Physical interactions between robots and humans are an integral part of many neurorobotic, neural prosthetic and rehabilitation robotics applications. It is generally acknowledged that such interactions can be enhanced by providing robots with advance knowledge of the intentions of human agents, e.g. their desired motor plans and goals in a given context. One potential source of these intentions are decoded neural signals obtained from the cerebral cortex, but precisely which cortical representations are most beneficial for facilitating effective human-robot interactions is unclear. Here we review the neural representations of movement plans in the cortex and discuss the potential utility of these representations for jointly performed motor actions, particularly manipulation tasks involving the hand and arm. Emphasis is placed on the coordinate frames used by different cortical areas to encode sensory-and motor-related variables. It is argued that relative coding of sensorimotor variables, a concept that has also recently been applied to robotic planning and control algorithms, might be particularly useful for facilitating joint actions of the hand and arm. More generally, discussion of the various neural representations will provide critical insight into how biological agents might better interact with robotic agents for the development of next-generation neural prosthetic systems and rehabilitation robots.