Multi-Sensor Control for Multi-Object Bayes Filters

摘要

Sensor management in multi-object stochastic systems is a theoretically andcomputationally challenging problem. This paper presents a novel approach tothe multi-target multi-sensor control problem within the partially observedMarkov decision process (POMDP) framework. We model the multi-object state as alabeled multi-Bernoulli random finite set (RFS), and use the labeledmulti-Bernoulli filter in conjunction with minimizing a task-driven controlobjective function: posterior expected error of cardinality and state (PEECS).A major contribution is a guided search for multi-dimensional optimization inthe multi-sensor control command space, using coordinate descent method. Inconjunction with the Generalized Covariance Intersection method formulti-sensor fusion, a fast multi-sensor algorithm is achieved. Numericalstudies are presented in several scenarios where numerous controllable (mobile)sensors track multiple moving targets with different levels of observability.The results show that our method works significantly faster than the approachtaken by a state of art method, with similar tracking errors.