Andrey Rudenko, Luigi Palmieri, and Kai O. Arras
Predictive Planning for a Mobile Robot in Human Environments
Proceedings of the Workshop on AI Planning and Robotics: Challenges and Methods (at ICRA 2017)
Abstract
Generating short and safe paths for mobile robots operating in crowded spaces is a hard task due to the high uncertainty of each person’s future behaviour. Classical path planning approaches often result in an over-constrained or overly cautious robot that fails to produce a feasible, safe path in the crowd, or plans a large, sub-optimal detour to avoid people in the scene. This work addresses these issues by exploiting long-term predictions of humans’ activities to plan short and safe paths for mobile robots in social spaces. We introduce a Markov Decision Process based motion prediction approach, which extends the baseline works by better adapting online to the observation of the pedestrian speed (policy cutting technique) and by introducing a fast random-walk based method (stochastic policy sampling) to generate predictions. Moreover, differently from the baselines, we evaluate several ways to incorporate predictions in path planning algorithms, and choose the most promising one. Through an extensive evaluation, using both simulated and real-world datasets, we show that our approach can accurately predict human motion and improve the quality of robot’s path planning.
@inproceedings{rudenko2017Predictions,
title={Predictive Planning for a Mobile Robot in Human Environments},
author={Rudenko, Andrey and Palmieri, Luigi and Arras, Kai Oliver},
booktitle={Proceedings of the Workshop on AI Planning and Robotics: Challenges and Methods (at ICRA 2017), Singapore},
year={2017}
}