Jack DiGiovanna

ETH Zurich, Switzerland

Embodied brain-machine interfaces: Past & Future

Summary

Brain-machine interfaces (BMI) include a wide variety of technologies which create bridges between the nervous system and an external machine. Ideally, BMI will provide a functional solution for persons afflicted with nervous system injuries and disorders. Additionally, BMI could revolutionize the way that any person interacts with tools. Research groups around the world have been advancing the state of the art in BMI over the past three decades and the technology is moving from animal models into clinical trials for humans. However, the technology is still imperfect and requires support from a researcher with advanced technical expertise on BMI architectures.

In this talk I present an overview of BMI architectures and the types of machine learning algorithms required to create the mapping between neural signals and machine control commands. Additionally, I draw parallels between the concept of embodiment in the artificial intelligence community and BMI design choices made by biomedical engineers and neuroscientists. Specifically, I review six major BMI implementations and show how the degree of embodiment in each affected performance. Finally, I propose a next generation of embodied BMI to improve performance and clinical acceptance. This new architecture creates a direct connection to a user’s spinal cord instead of to an external prosthetic. The computational learning agent in this architecture is embodied because it does not interact directly with the environment but instead learns to exploit intact spinal circuits to perform complex behaviors and process sensory feedback.