Heterogeneous Multi-Robot Person-Following in Constrained Environments

There are many reasons for a robot to track a moving person, from security monitoring to providing social assistance. In this work, we consider the problem of autonomously tracking a moving person through an altitude-constrained environment using mobile robotic platforms. This problem presents several challenges. Because we are constrained to low altitudes, such as in indoor environments or below the roof or treeline in an outdoor environment, in order to maintain sensor tracking the platforms must strongly consider obstacles and occlusion. Because we consider the robot's maximum speed to be constrained to the same order of magnitude as the target, the robots must make timely and correct navigation decisions to maintain track. We examine solutions to these challenges using a robot team consisting of a slower, more-capable ground robot and a faster, less-capable micro aerial vehicle. Three methods for robot navigation decision-making are evaluated: a heuristic-based predictive goal search algorithm, a multi-agent reinforcement learning method, and a baseline closed-loop method. While the heuristic method proved the most reliable, completing all nine paths and reaching targets nearly twice as fast as RL, we were still able to demonstrate a successful MARL-based solution. These findings reveal both the potential and limitations of RL, demonstrating the promise of heterogeneous multi-robot systems for person-following in unpredictable settings.