Adaptive Online Distributed Optimal Control of Very-Large-Scale Robotic Systems

Abstract

Autonomous systems comprised of many cooperative agents have the potential for enabling long-duration tasks and data collection critical to the understanding of a wide range of phenomena in spatially and temporally variable environments. The adaptive distributed optimal control approach presented in this article extends online approximate dynamic programming to very-large-scale robotics (VLSR) systems that must operate and adapt to highly uncertain and variable environments. Optimal mass transport theory is used to show that, in the Wasserstein-Gaussian mixture model space, the VLSR system's cost to go can be represented by a value functional of the robot distribution and dynamic environmental maps. The approach is demonstrated on a cooperative path planning problem in which knowledge of the obstacles in the environment changes incrementally over time based on in situ measurements. Numerical simulations show that the proposed approach significantly outperforms existing methods by finding an approximately optimal solution that avoids obstacles and meets a desired final robot distribution using minimum energy.