Poster: Two-Phase KalmanNet Based Collaborative Detection Framework
Tianlong Zhou, Weixiong Rao, Feng Ye · 2023
The collaborative detection problem has been widely used in many applications. Existing works typically exploit a Kalman Filter or its variant to estimate target state with an impractical assumption that the state space and environment are fully known. To address this issue, we propose a novel multi-agent reinforcement learning (MARL) based collaborative detection framework. The key is (1) a Two-Phase Kalman Neural Network (TPKN) to estimate target state and (2) a reinforcement learning (RL) model by taking the target estimation state as input and generating an action to track targets. Our initial evaluation with 4 pursuer agents and 4 targets demonstrates that our framework outperforms the state-of-the-art by a much higher tracking ability and lower localization error.11Weixiong Rao and Feng Ye are joint corresponding authors of the paper.