Interactive Route-Planning and Mobile Sensing with a Team of Robotic Vehicles in an Unknown Environment

Abstract

View Video Presentation: https://doi.org/10.2514/6.2021-0865.vid We study the problem of decentralized route-planning of a team of multiple mobile robotic actors in an unknown environment. Mobile sensors such as aerial surveillance vehicles can be deployed to learn the environment map, which leads to a sensor placement problem. A traditional solution to this problem is to decouple sensor placement and route-planning. We show that this decoupled approach can be wasteful in sensory resources (numbers of sensors and measurements). Instead, we propose a bootstrapping, iterative, and interactive route-planning and sensor placement technique that finds near-optimal routes for actors that are required to work collaboratively to satisfy a global task. To this end, a task-driven information gain is defined based on entropy reduction over a subregion “near” current estimates of optimal routes. The crucial innovation is that the proposed sensor placement attempts to reduce the entropy of the actors’ route cost rather than that of the entire map. Sensors are placed to maximize the task-driven information gain. The iterations of route-planning and sensor placement terminate when the entropy of each actor’s route cost reduces below a desired threshold. Numerical simulations demonstrate significant reductions in the number of measurements required to find near-optimal actor routes.