An Animal-in-the-Loop, Approximate Dynamic Programming Based Robotic Design Paradigm

Arizona State University, Scottsdale AZ

Investigators

Abstract

The future development of space solar power systems presents immense challenges for designers of sophisticated and adaptable robotic systems capable of complex mission assignments in the difficult environment of space. In an effort to help address a subset of these problems, this project will conduct an interdisciplinary investigation integrating the diverse ranges of computer science, neurophysiology, psychology and bioengineering. It will develop a novel animal-in-the-loop robotic architecture for the efficient investigation and optimization of learning algorithms applicable to general cooperative robotic systems. The plan is to eclipse the current trend for bio-inspired robotics through the use of an actual bio-based robot. This will be accomplished by training rats to pilot mobile robots. Rats will be used in place of the traditional sensing mechanisms of robotic systems. In so doing, the researchers aim to exploit the proven superiority of embedded processing paradigms provided by real neural networks. To better understand exactly how and why mammals are such good processors of sensory input and how we might apply their strategies to robotic systems, the researchers will equip each rat with multi-electrode arrays in the sensory regions of their brain. They will collect neural spike data representing the sensory environment as the rat-robot systems complete a cooperative task. Task data and neural recordings will be used in the development and simulation of advanced learning algorithms based on approximate dynamic programming (ADP). A better understanding of sensory processing coupled with advances in ADP models will be used to design more intelligent platforms, namely animal-computer co-adaptation, in cooperative robotic systems. For the development of space solar power objectives, such robotic systems must provide the essential traits of robustness and scalability. The new generalized approach using bio-sensor based, ADP optimized solutins provides a unique high-potential to develop such powerful systems.