Probabilistic Approach to Robot Motion Planning in Dynamic Environments

Abstract

Four major approaches to robot motion planning in dynamic environments are discussed: probabilistic robot, probabilistic collision state (PCS), partially closed-loop receding horizon control (PCLRHC) and gross hidden Markov model (GHMM). A comparison of three mapping techniques, Kalman filter, expectation and maximization algorithm and Markov model, is presented. The PCS method is the probabilistic extension of inevitable collision state, which is found to be the safest motion planning method. The concept of open-loop and partially closed-loop receding horizon control (OLRHC and PCLRHC) is compared critically, and the algorithms are benchmarked. GHMM is the best suited method for environments with limited space and dynamic environment due to human interactions. GHMM parameters and structure are evaluated using an incremental “learn-and-predict” approach. For exploring GHMM, we simulated a cafeteria with eight tables to be served by a robot, considering three different arrangements of tables along with convex and concave obstacles, and obtained the path length and time taken for a Hamiltonian path. During the simulation, it was observed that for a given static or dynamic environment, the concavity of the obstacles is what makes the scenario a complex one.