Abstract
To make robots coexist and share the environments with humans, robots should understand the behaviors or the intentions of humans and further predict their motions. In this paper, an A*-based predictive motion planner is represented for navigation tasks. A generalized pedestrian motion model is proposed and trained by the statistical learning method. To deal with the uncertainty, a localization, tracking and prediction framework is also introduced. The corresponding recursive Bayesian formula represented as DBNs (Dynamic Bayesian Networks) is derived for real time operation. Finally, the simulations and experiments are shown to validate the idea of this paper.
Highlights
By the efforts of robotic researchers, there has been a great progress in robotic techniques
According to Eq(1), the pedestrian model is factorized into motion model in distance map (DistMap) and navigation function (NF)
We proposed a predictive navigation planner for a mobile robot in populated environments
Summary
By the efforts of robotic researchers, there has been a great progress in robotic techniques. Lots of researchers developed efficient replanning algorithms to cope with the real time challenge [9, 18, 29, 38]. When the planning problem is complex, a complete and optimal path may not be found within the limited deliberation time. Anytime algorithms are useful and have been shown the excellent results in this situation [14, 19] It generates a suboptimal path quickly in the beginning and further improves the path until the deliberation time has run out. In this paper, the proposed planning algorithm is not addressed in the POMDPs framework. A predictive motion planner is developed for dynamic environments.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
