International Conference on Control, Automation and Systems 2007 Oct. 17-20,2007 in COEX, Seoul, Korea Safe steering of UGVs in polygonal environments Yongsoon YoonI, Tokson Choe", Yongwoon Park", and H. Jin Kim I 1 School of Mechanical & Aerospace Eng. and Inst. of Advanced Aerospace Technology Seoul National University, Seoul, 151-742, KOREA Tel: +82-2-880-7392, Email: {ilssoon6,hjinkim}@snu.ac.kr 2 Agency for Defense Development, Jochiwongil461, Yuseong, Daejon, KOREA Tel: +82-42-822-4271, Email: tschoe@add.re.kr, woon5901 @hanafos.com Abstract: This paper presents an application of a model predictive control for trajectory generation of an unmanned ground vehicle. An optimal tracking problem while avoiding collision with obstacles is formulated in terms of cost min- imization under constraints. The cost function includes terms corresponding to the deviation from the desired trajectory, magnitude of the control input, proximity to the obstacles and the final destination point, respectively. Information on obstacles can be incorporated online in the nonlinear model predictive framework and the resulting constrained optimiza- tion problem can be solved using nonlinear programming techniques such as augmented Lagrangian. Then kinematic constraints are treated by the Karush- Kuhn-Tucker (KKT) condition. This approach has been applied for generating safe trajectories for the nonlinear dynamics of a vehicle with a nonlinear tire model in a 2D polygonal environment. Simu- lation results show that the satisfactory performance was achieved in terms of short and safe trajectory satisfying input constraints. Keywords: Model Predictive Control, Obstacle Avoidance, Unmanned Ground Vehicle 1. INTRODUCTION Recently, unmanned ground vehicles (UGVs) have been the topic of much research and development due to their applicability in various fields. Many of these appli- cations require UGVs to move in unknown environments with dynamic and physical constraints. Often UGVs are commanded to perform pre-defined maneuvers or to fol- Iowa pre-planned path designated by an off-line mission- level planning algorithm. However, with unidentified ob- stacles due to sensor limitations or erroneous a priori information, performing collision-free maneuvers is not easy. Many researches on local obstacle avoidance, which use purely reactive methods based on sensor input, have been performed [1], [2], [3], [4]. Some take into account the dynamics and kinematics constraints[5]. These ap- proaches are computationally efficient, but the vehicle can get stuck in local minima and sometimes the dis- cretization of the world is required or the full dynamics cannot be incorporated. Moreover at high speeds, obsta- cle avoidance becomes harder with this reactive methods in spite of a wide sensor range. Despite powerful on- board computers, at high speeds there is little time to per- form replanning based on detailed vehicle models. Recently, predictive active steering control for au- tonomous vehicle systems was studied [7], [8], [9] with a nonlinear tire model [6]. In these works, the autonomous vehicle was directed to follow the given reference which is assumed to be collision-free and achievable. NPSOL and QP solvers were used to solve the optimization prob- lem with the nonlinear dynamics. Often only the trajec- tory of center of gravity of the vehicle was considered. But sometimes we have to consider the dimension of the vehicle. If there exist obstacles of various shape, the di- mension of the vehicle is critical for collision-free move- Fig. 1 The UGV in consideration, its mass, momentum of inertia, length and width are 20 kg, 1.0542 kg-rn", 0.6 m and 0.32 m, respectively. ment. In multi-vehicle system since UGV model con- sidered with tire models is nonholonomic and the envi- ronment changes very dynamically, collision avoidance is not easy. It may fail due to a short look-ahead horizon or a lack of information shared among the vehicles. In this paper, we present a model predictive method for active steering control of UGVs based on successive on- line optimization. In order to use this approach as a local obstacle avoidance planner for UGVs, we use a bicycle model to predict the future evolution of the system. We consider the collision avoidance as well as trajectory fol- lowing with considerations of dimension of the UGVs, shape of the obstacles and different levels of communi- cation capability between the UGVs. If a vehicle runs into stationary or dynamic obstacles on the way to the target point, controller predicts a future path and solves an optimization problem to replan collision-free trajec- tories without deviating too much from the pre-defined reference trajectory. Nonlinear model predictive control has been used to generate safe trajectories for unmanned aerial vehicles [10], but without including the saturation limits as constraints. In the current paper, safety limits 978-89-950038-6-2-98560/07/$15 @ICROS 1804