Policy Learning for Active Target Tracking over Continuous SE(3) Trajectories

TL;DR

This paper introduces a model-based policy gradient method for active target tracking with a mobile robot in continuous 3D space, optimizing sensor measurements to reduce target uncertainty.

Contribution

It presents a neural network control policy that incorporates robot pose and target distribution info, with an explicit gradient derivation for efficient optimization.

Findings

Effective reduction of target distribution entropy achieved

Neural network policy handles variable number of targets

Explicit gradient derivation improves training efficiency

Abstract

This paper proposes a novel model-based policy gradient algorithm for tracking dynamic targets using a mobile robot, equipped with an onboard sensor with limited field of view. The task is to obtain a continuous control policy for the mobile robot to collect sensor measurements that reduce uncertainty in the target states, measured by the target distribution entropy. We design a neural network control policy with the robot $SE(3)$ pose and the mean vector and information matrix of the joint target distribution as inputs and attention layers to handle variable numbers of targets. We also derive the gradient of the target entropy with respect to the network parameters explicitly, allowing efficient model-based policy gradient optimization.