Policy Learning for Active Target Tracking over Continuous SE(3) Trajectories
This work addresses active target tracking for robotics applications, presenting an incremental improvement by integrating neural networks with model-based gradients for SE(3) trajectories.
The paper tackles the problem of tracking dynamic targets with a mobile robot using a limited field-of-view sensor, by proposing a model-based policy gradient algorithm that learns continuous control policies to reduce target state uncertainty, measured by entropy, and demonstrates efficient optimization through explicit gradient derivation.
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.