Motion Planning Method Based On Trajectory Primitives And Its Applications In Autonomous Land Vehicles

During my study for PHD, I have participated in two projects of"the 10th Five-Year Plans of Advanced Defense Technologies"and been responsible for the motion control of one project and the motion planning of the other project. Although these projects have been confirmed and accepted, some key technologies will be investigated continually in the following projects of"the 11th Five-Year Plans of Advanced Defense Technologies". Cross-country motion planning plays a key role in the development of unmanned armed vehicles. Yet its adaptability to complicated environments is to be improved further.The main task of the motion control is to regulate the velocity and direction of the vehicle and keep the vehicle's position and attitude consistent with the planned trajectory. In traditional motion planning algorithms, only geometric constraints are considered, while the kinematics and dynamics of the vehicle are ignored. Therefore the planned trajectory is not guaranteed for feasibility. That is, sometimes motion control cannot make the vehicle precisely track the planned trajectory. Since traditional motion planning algorithms are based on the decomposition of the configuration space, it is difficult to incorporate the kinematics and dynamics of the vehicle into the old framework. Generally the parameters of the trajectory, such as the maximum curvature and the maximum curvature rate, are not modified iteratively untill all obstacles are avoided successfully. This"try and error"method severely tempers the maneuverability of the vehicle in complex environments. It is impossible to guarantee the feasibility of the planned trajectory and improve the trajectory via the incorporation of dynamic features of the vehicle. My experience with the projects suggests that motion planning and motion control should couple tightly. It is in the conjunction that can we find a good method to tackle the motion-planning problem in off-road environments.In this dissertation, the problems of current motion-planning algorithms are analysed. The design of the vehicle's trajectory is based on the kinematics and dynamics of the vehicle, instead of the decomposition of the configuration space. It shows that trajectory primitives implicate the dynamic properties of the vehicle and can be used to generate the expected trajectory through connections. Which trajectory can be used and how to connect them preserving the kinematic and dynamic constraints are the main focuses of this dissertation. Traditional methods used in engineering are reviewed briefly, just to explain the differences between the new algorithm and the old ones. The results and innovations in this dissertation can be summarized as follows.Above all, a new motion planning method called motion planning based on trajectory primitives is proposed. A real-time cross-country motion planning system has been developed in a vehicle platform.Secondly, it is for the first time that a formal language is given to describe the construction and connection of trajectory primitives. Furthermore, necessary conditions of the formal system's controllability are proven and then used in the following design of trajectory primitives. In this thesis, the dynamic equations of mechanic systems are viewed from the perspective of Lie group. The new method is proposed based on the two dynamical properties of symmetry and relative equilibrium, which are owned by general mechanical systems. Trajectory primitives consist of trim trajectories and maneuver trajectories. Trim trajectories are derived from relative equilibriums, while maneuver trajectories are given when the system states jump from one relative equilibrium to the other.Thirdly, from the perspective of engineering, the selection of the trim primitives is solved and then the second-order reduced dynamics of the dynamic model of the vehicles is presented. A design method of the maneuver primitives, which is used to connect the trim primitives, is proposed based on optimal Bangbang principle.Fourthly, dynamic programming is used to implement optimal motion planning based on trajectory primitives. Moreover, randomized sample planning is used to realize rapid motion planning based on trajectory primitives.Ground tests for the projects of"the 10th Five-Year Plans of Advanced Defense Technologies"had proved the adaptiveness of this method in different scenarios.