.png){kind=icon}
Return
A Robust Sensor Scheduling Algorithm Based on Deep Reinforcement Learning for Maneuvering Target Tracking
.png){kind=icon}
PRE
AI
DOI：10.1109/jiot.2026.3719505.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
This work considers the problem of sensor scheduling for maneuvering targets using time difference of arrival (TDOA). We propose a novel interacting multiple model (IMM) sensor scheduling method based on the parameterized deep Q-network (PDQN) algorithm. This method uses extended Kalman filtering (EKF) to predict the trajectory of the maneuvering target. Then, PDQN is utilized to optimize sensor scheduling, thereby improving the applicability of maneuvering target tracking. In this framework, we design an innovative reward function that integrates the tracking uncertainty characterized by the Cramér–Rao lower bound (CRLB) with the physical sensor constraints, guiding the agent to learn a strategy that maximizes long-term tracking accuracy. To ensure robustness to the target motion patterns, a domain-randomization strategy is adopted during training. A large number of numerical experiments in various scenarios show that the proposed method has comparable tracking accuracy to the optimization-based control benchmark, and achieves better performance in some cases while reducing the online decision-making time by an order of magnitude. Our work establishes deep reinforcement learning (DRL) as an effective alternative for solving complex joint estimation and control problems, especially in scenarios that require long-term planning and high robustness to uncertainties.
Keywords:
Cramér–Rao lower bound (CRLB)
extended Kalman filtering (EKF)
interacting multiple model (IMM)
parameterized deep Q-network (PDQN)
sensor scheduling
target tracking
Journal
IF:
8.9
Papers:
1.4W
Citations:
7.8W
Organization
Cited Papers
No cited papers available