Obstacle Avoidance for Wheeled Robots in Unknown Environments using Model Predictive Control

Abstract

This paper presents a model predictive approach for obstacle avoidance of car-like unmanned ground vehicles (UGVs). An optimal tracking problem while avoiding obstacles in unknown environments is formulated in terms of cost minimization under constraints. Information on obstacles can be incorporated online in the nonlinear model predictive framework and kinematic constraints are treated by Karush-Kuhn-Tucker (KKT) condition. The overall problem is solved real-time with nonlinear programming. This approach is applied to car-like robots including tire models while explicitly considering the dimension of the UGV rather than treating it as a dimensionless cart model. Two kinds of potential-like terms are employed in the cost function for obstacles avoidance. The first term is to consider the distance between the UGV and the obstacle, and the second one is to consider the parallax information of the UGV about the obstacles. Simulation results show that both two approaches can make safe steering in a simple environment, but in a complex environment such as an urban area, the approach based on the modified parallax (MP) was more successful in the view of the computation time and safe steering.