Abstract:

A path planning algorithm, named “multi-vehicle Monte Carlo Bayes filter tree (MV-MCBFT)”, was proposed for the distributed active perception of multiple autonomous vehicles to estimate the occupancy status of parking lots in real time. A sequential, feed-forward cooperative motion planning strategy driven by predicted observations was proposed by constructing a probabilistic state-transition model for parking lots, with designing a multi-source Bayesian filtering fusion mechanism, and with incorporating submodular maximization principles. The results show that the MV-MCBFT achieves near-optimal performance consistent with the traversal algorithm in terms of entropy reduction ratio and estimation accuracy, while consuming only 1% of the runtime required by the traversal algorithm. The MV-MCBFT has the entropy reduction ratio by 43.70% and the estimation accuracy by 51.43% comparing with the random-walk algorithm. Therefore, the proposed method enhances the effectiveness of parking lot state estimation.

Key words: autonomous vehicles, informative path planning, Bayes filter, parking occupancy status estimation