Experimental Performance Analysis of a Homotopy-Based Shared Autonomy Framework

This paper describes and experimentally demonstrates a new approach to shared-adaptive control of human-machine systems. Motivated by observed human proclivity toward fields of safe travel rather than specific trajectories, our approach is rooted in the planning and enforcement of constraints rather than the more traditional reference paths. This approach identifies path homotopies, bounds a desired homotopy with constraints, and allocates control as necessary to ensure that these constraints remain satisfied without unduly restricting the human operator. We present a summary of this framework's technical background and analyze its effect both with and without driver feedback on the performance and confidence of 20 different drivers teleoperating an unmanned (teleoperated) vehicle through an outdoor obstacle course. In 1200 trials, constraint-based semiautonomy was shown to increase the operator speed by 26% while reducing the occurrence of collisions by 78%, and improving overall user confidence and sense of control by 44% and 12%, respectively-all the while assuming less than 43% control of the vehicle.