Generating informative paths for persistent sensing in unknown environments

Abstract

In this thesis, we present an adaptive control law for a team of robots to shape their paths to locally optimal configurations for sensing an unknown dynamical environment. As the robots travels through their paths, they identify the areas where the environment is dynamic and shape their paths to sense these areas. A Lyapunov-like stability proof is used to show that, under the proposed adaptive control law, the paths converge to locally optimal configurations according to a Voronoi-based coverage task, i.e. informative paths. The problem is first treated for a single robot and then extended to multiple robots. Additionally, the controllers for both the single-robot and the multi-robot case are extended to treat the problem of generating informative paths for persistent sensing tasks. Persistent sensing tasks are concerned with controlling the trajectories of mobile robots to act in a growing field in the environment in a way that guarantees that the field remains bounded for all time. The extended informative path controllers are proven to shape the paths into informative paths that are useful for performing persistent sensing tasks. Lastly, prior work in persistent sensing tasks only considered robotic systems with collision-free paths. In this thesis we also describe a solution to multi-robot persistent sensing, where robots have intersecting trajectories. We develop collision and deadlock avoidance algorithms and quantify the impact of avoiding collision on the overall stability of the persistent sensing task. Simulated and experimental results support the proposed approach.