Dynamic Obstacle Avoidance for an Omnidirectional Mobile Robot

We have established a novel method of obstacle-avoidance motion planning for mobile robots in dynamic environments, wherein the obstacles are moving with general velocities and accelerations, and their motion profiles are not preknown. A hybrid system is presented in which a global deliberate approach is applied to determine the motion in the desired path line (DPL), and a local reactive approach is used for moving obstacle avoidance. A machine vision system is required to sense obstacle motion. Through theoretical analysis, simulation, and experimental validation applied to the Ohio University RoboCup robot, we show the method is effective to avoid collisions with moving obstacles in a dynamic environment. 1. Introduction An omnidirectional robot is a holonomic robot that can move simultaneously in rotation and translation [1]. Most work on omnidirectional robots is in robot development; the few studies on dynamic models are Watanabe et al. [2], Moore and Flann [3], Williams et al. [4], and Kalmar-Nagy et al. [5]. These models all have decoupling between the wheels, which is not complete; thus, we first briefly summarize a new coupled nonlinear dynamics model for three-wheeled omnidirectional robots. The potential field method was first suggested by Andrews and Hogan [6] and Khatib [7] for obstacle avoidance of manipulators and mobile robots. Obstacles exert a virtual repulsive force, while the goal applies a virtual attractive force to the robot. Koren and Borenstein [8] identify potential field limitations (robot trapped by local minima, oscillation in presence of obstacle, and the lack of passage between closely spaced obstacles). To overcome these problems, they developed the vector field histogram. Ge and Cui [9] mentioned an additional shortcoming, a nonreachable goal with an obstacle nearby, and presented a new repulsive function to overcome it, increasing complexity and computation. Adams [10] presented a simulation study using the potential field method considering low-level robot dynamics, with static obstacles. Guldner and Utkin [11] proposed a method that took the gradient of the potential field as the desired vector field for path planning. Tsourveloudis et al. [12] proposed an electrostatic potential field for an autonomous mobile robot in a planar dynamic environment; their method depends on obstacle prediction accuracy and slow environment changes. The velocity space method to deal with moving obstacle avoidance in a preknown environment was suggested by Fiorini and Shiller [13], who discussed the velocity obstacle concept. This method