Distributed multi-robot exploration under complex constraints

Abstract

Mobile robots have emerged as a prime alternative to explore physical processes of interest. This is particularly relevant in situations that have a high risk for humans, like e.g. in search and rescue missions, and for applications in which it is desirable to reduce the required time and manpower to gather information, like e.g. for environmental analysis. In such context, exploration tasks can clearly benefit from multi-robot coordination. In particular, distributed multi-robot coordination strategies offer enormous advantages in terms of both system’s efficiency and robustness, compared to single-robot systems. However, most state-of-the-art strategies employ discretization of robots’ state and action spaces. This makes them computationally intractable for robots with complex dynamics, and limits their generality. Moreover, most strategies cannot handle complex inter-robot constraints like e.g. communication constraints. The goal of this thesis is to develop a distributed multi-robot exploration algorithm that tackles the two aforementioned issues. To achieve this goal we first propose a single-robot myopic approach, in which we build to develop a non-myopic informative path planner. In a second step, we extend our non-myopic single-robot algorithm to the multi-robot case. Our proposed algorithms build on the following techniques: (i) Gaussian Processes (GPs) to model the spatial dependencies of a physical process of interest, (ii) sampling-based planners to calculate feasible paths; (iii) information metrics to guide robots towards informative locations; and (iv) distributed constraint optimization techniques for multi-robot coordination. We validated our proposed algorithms in simulations and experiments. Specifically, we carried out the following experiments: mapping of a magnetic field with a ground-based robot, mapping of a terrain profile with two quadcopters equipped with an ultrasound sensor, and exploration of a simulated wind field with three quadcopters. Results demonstrate the effectiveness of our approach to perform exploration tasks under complex constraints.