Home research People General Info Seminars Resources Intranet
Abstract

Ali-akbar Agha-mohammadi, Saurav Agarwal, Aditya Mahadevan, Suman Chakravorty, Daniel Tomkins, Jory Denny, Nancy M. Amato, "Robust Online Belief Space Planning in Changing Environments: Application to Physical Mobile Robots," Technical Report, TR13-007, Parasol Laboratory, Department of Computer Science, Texas A&M University, Jul 2013.
Technical Report(pdf, abstract)

Methods based on the POMDP (Partially-Observable Markov Decision Process) framework for planning under uncertainty rely on the knowledge about the system model and the noise models. However, in practical cases there are always discrepancies between the models used for computation and the realworld models. This paper proposes a dynamic replanning scheme which can perform real-time replanning as the system deviates from the desired plan or encounters any sign of such discrepancies. The main contribution of this paper is to implement a belief space planner on a physical robot and making POMDP methods one step closer to becoming a practical tool for robot motion planning. We demonstrate the planning results on an i-robot create equipped with a monocular camera. Moreover, we show that not only is the proposed method robust to model discrepancies, but also that it is robust to changes in the environment (both in the obstacle map and information sources), as well as unforeseen large deviations in robot’s location.