Policy Gradient for Observer Trajectory Planning with Application in Multi-target Tracking Problems

Tracking multiple moving targets with bearing-only measurement is a challenging task, due to the inherent difficulties in determining the correct trajectory of the observer that will meet observability conditions. The work presented here formulates Observer Trajectory Planning (OTP) as a continuous control problem, and proposes reinforcement learning as a solution. The proposed architecture in this work constitutes a model-independent framework that allows for the estimation of the states of targets, and that allows multiple targets to be tracked in a realistic scenario, where the agent has no prior information about the initial locations and velocities of the targets.