Overview
Many surveillance systems are tasked with monitoring the area of interest and tracking targets of interest by estimating their pose and velocity over a period of time. Target tracking is challenging because these sensor measurements can be intermittent and noisy depending on the applications and environmental conditions, and the true target state is never available. Therefore, formulating and evaluating the uncertainty is important in target tracking to assess the system performance and to make further decisions. In this project, we aim to develop information-driven trajectory planning approaches for multi-agent multi-target tracking under intermittent measurements and target motion model uncertainty from the trained neural network.
Research Questions
- How can we estimate information gains in continuous domain for a long-horizon planning?
- How can an information-driven approach represent sensor measurements utility considering motion model learning?
- How can we compute expected information gain for multiple agents for multi-target tracking?
Research Team
PI: Jane Shin
Participants: Andres Pulido, Kyle Volle, Kristy Waters, Jared Paquet
PO: Zach Bell
Acknowledgement
This work is supported by Air Force Research Lab through FA8651-22-S-0001.