REAL-TIME PATH PLANNING AND TRACKING CONTROL USING A NEURAL DYNAMICS BASED APPROACH ⋆ Simon X. Yang and Eric Hu Advanced Robotics and Intelligent Systems (ARIS) Lab School of Engineering, University of Guelph Guelph, ON N1G 2W1, Canada Abstract: Real-time collision-free path planning and tracking control of a nonholo- nomic mobile robot in a dynamic environment is investigated using a neural dynamics based approach. The real-time robot path is generated through a dynamic neural activity landscape of a topologically organized neural network that represents the changing environment. The dynamics of each neuron is characterized by an additive neural dynamics model. The real-time tracking velocities are generated by a novel non-time based controller, which is based on the conventional event based control technique and an additive model. The effectiveness and efficiency of this approach are demonstrated through simulation and comparison studies. Copyright c 2002 IFAC Keywords: mobile robots, path planning, obstacle avoidance, velocity control, neural dynamics, neural network 1. INTRODUCTION Real-time collision-free path planning and track- ing control of mobile robots are fundamentally im- portant but very difficult issues in robotics, partic- ularly in a nonstationary environment. Many pre- vious works deal with the path planning and the tracking control separately. (e.g. Zelinsky, 1994; Jiang et al., 1997; Podsedkowski, 1998; Svestka and Overmars, 1997; Xi, 1993; Kang et al., 1999). There are many studies on robot path planning using various approaches. Most of the previous models for path planning deal with static environ- ment only and are computationally complicated. A local collision checking procedure is required at each step of the robot movement, e.g., to de- tect local collisions, Zelinsky (1994) model uses a hierarchical collision testing procedure based on “distance space bubbles”. ⋆ This work was partially supported by Natural Sciences and Engineering Research Council (NSERC) and Materials and Manufacturing Ontario (MMO) of Canada. Most previous models for nonholonomic mobile robots use two-step approaches that first compute a collision-free holonomic path, and then trans- form this path by a sequence of feasible ones. The quality of the solution and the computational cost of the second step depend on the shape of the holonomic path, e.g., Jiang et al. (1997) proposed a time-optimal path planning method for a robot with kinematic constraints, which consists of three stages: planning for a point mobile robot; planning for a mobile robot; and optimizing cost functions for a time-optimal solution. Podsedkowski (1998) proposed a path planner for nonholonomic mobile robot using a searching algorithm, which requires a local collision checking procedure and the mini- mization of cost functions. There are some learning based path planning models, e.g., Svestka and Overmars (1997) pro- posed a probabilistic learning approach to path planning of mobile robots, which involves a learn- ing phase and a query phase and uses a local method to compute the feasible paths for the Copyright © 2002 IFAC 15th Triennial World Congress, Barcelona, Spain