On-line Real-time Path Planning Of Mobile Robots In Dynamic Uncertain Environment

Path planning of mobile robots is one of key issues in robotics research. It is always a focus topic for the researchers all over the world. However, traditional methods of path planning, whatever global methods or local methods, all have their own drawbacks respectively. Global planning methods can give optimal planning results. But they can only plan once off-line so that real-time capability cannot be executed. While local planning methods can implement real-time planning. But since there is no global information included, such indices as motion path or runtime cannot be optimized. With more and more complex work environment and higher demands of the planning tasks, robots will not complete the tasks well if traditional methods of path planning are still adopted.To the above drawbacks, combining the global planning with the local planning, this dissertation presents a new approach to on-line real-time path planning of mobile robots based on polar coordinates space in which the desirable direction angle is taken into consideration as an index of path optimization. Detecting unknown obstacles with local feedback information by the robot's sensor system, the approach orients the desirable direction of the mobile robot so as to generate local sub-goal in every planning window. As a result, the difference between real direction angle and desirable direction angle of robot motion steers the mobile robot to detour collisions and advance towards the target without stopping to re-plan a path when new sensor data become available. Moreover, this approach is applied to the path planning of mobile robots in dynamic uncertain environment. The theoretic issues of the planning algorithm, such as astringency, security and reach-ability, are lucubrated in the dissertation. The effectiveness, feasibility, high stability, real-time capability and perfect performance of obstacle avoidance are demonstrated by simulation examples and the experiment. The innovations of this approach are shown as follows:1. The desirable direction angle taken into consideration as a path optimization index is completely different from those traditional methods of path planning based on C-space and Cartesian coordinates space. It does not concern how long the robot moves but in what direction it moves, that is, determining desirable direction of the robot's motion is just the task in each local planning. In every planning window, the desirable direction angle is always determined on the basis of the global information—the goal. So the optimal planning to the robot's motion path can be implemented in each local planning.2. Polar coordinates are presented as the local coordinates of the robot. In one hand, we select 2D laser ranging radar as the main sensor of the robot. It scans the surrounding environment with certain angular discrimination in its scan plane. And its original data are all the discrete data. Each point is expressed by the distance p and corresponding scan angle a , that is, the polarform (p,a). So the decision of the desirable direction angle is very simple and easy toimplement. In another hand, adopting polar coordinates space not only can visually express the motion direction angle of the robot but also can be convenient for the computation of the motion direction angle so that planning time can be greatly saved. Consequently, real-time capability of path planning can be guaranteed.3. The path planning approach presented in this dissertation is executed with the organic integration of global information and local information. In each sampling period, the planning window always includes the global information—the goal. And the sub-goal of each local planning can be generated with the local environment information detected by the sensors. Adopting such idea, the approach not only can optimize the motion path for the robot but also is not easy to get into the local extrema like many traditional methods of local planning, that is, the deadlock phenomenon.4. With respect to the dynamic uncertain environment, the motion trajectories of moving obstacles are predicted by using an autoregressive (AR) model in this dissertation to enable the robot give a quick response when encountering obstacles. In each sampling period, positions of moving obstacles are sampled by the robot's sensor system. With current sampling positions, the AR model predicts future positions of moving obstacles in the next sampling period. And the predicted positions are treated as instantaneously static. So moving obstacles in the predicted positions can be considered as static obstacles in the planning process. Then the robot's collision-free path is planned with such "static" obstacles. Thus, dynamic path planning is translated into instantaneously static one.