CAREER: Preventive Robotics: Learning and Adaptation for Predictive Human Robot Symbiosis

Arizona State University, Scottsdale AZ

Investigators

Abstract

Deploying assistive technologies that intelligently minimize the risk of musculoskeletal injury during physical tasks could improve user safety and significantly reduce the healthcare costs associated with the treatment of long-term disabilities such as chronic pain. With that goal in mind, this CAREER project will contribute key innovations that allow robots to reason about the biomechanical safety of actions performed jointly with a human partner. The research will focus on the concept of Preventive Robotics, a novel approach to human-machine collaboration that incorporates the biomechanical well-being of the human user into robot control and decision-making. In contrast to Rehabilitation Robotics, which focuses on therapeutic procedures after an injury occurs, Preventive Robotics seeks to proactively reduce the risk of injury. A critical knowledge gap in this regard is the absence of a theoretical foundation that supports human-machine symbiosis - healthy, physical, and bi-directional interactions between human and machine which can be comfortably sustained over very long periods of time. The main objective of Preventive Robotics is to generate assistive robot actions that (a) seamlessly blend with actions of the human partner to achieve the intended function, while (b) minimizing biomechanical stress on the human body. Coalescing these two goals will unlock new potential for robotics to drastically improve public and occupational health. The project will also involve transition of innovations to a commercial partner developing intelligent lower-leg prostheses. The research integrates with an education program targeting K-12 students, undergraduate and graduate students, and students from underrepresented groups. To these ends, the project will develop a unified Bayesian framework for modeling symbiotic dynamics among multiple agents using a compact probabilistic and data-driven methodology. The framework will bridge the divide between predictive modeling of humans and predictive control of symbiotic human-robot systems. A Bayesian representation will be used to derive algorithms for learning and adaptation which include the future biomechanical state of a human user. In addition, new symbiotic control algorithms will be introduced that utilize predicted biomechanical variables to steer the human-robot interaction towards biomechanically safe movement regimes. These control methods will provide new insights about strongly-coupled systems with reciprocal dependencies, in which only one system can be actively controlled (e.g., an assistive device or prosthesis). The new approach will be implemented on a powered-ankle prosthesis in order to anticipate joint loads and proactively avoid high stresses. The resulting prosthesis will have the potential to significantly lower the risk of musculoskeletal diseases such as osteoarthritis. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.