| Exploration planning is one of the most important issues for automatic mapping in unknown environment of mobile robot.It has significant theoretical and practical influence on autonomy,effectiveness,robustness and accuracy of mapping.The main target of exploration planning is to generate real-time motion control for robots that enlarges the perceptive range of the robot to cover largest area in a shortest time.Because the map information is incompleted,it is a big challenge to find an optimal and complete path in real time.Single-step planning,multi-steps planning and cooperative exploration of multi-robots are studied in this dissertation. The main work is as follows:①A robot simulation system based on Microsoft Visual C++ and Open Dynamics Engine(ODE) is designed to construct the virtual indoor static and dynamic obstacles,provide the virtual measurement of laser range finder and odometer,and simulate the kinetics motion of robots and obstacles.Meanwhile,the system could become a real-time platform for the researches on mapping, exploration planning,localization,navigation and multi-robots cooperation.②A fuzzy evaluation based exploration planning method is presented to deal with the fuzziness and uncertainty of unknown environmental information.Frontier points are classified by their discriminative distance and feasibility.With classification results,the candidate points with higher priority are evaluated according to their distance,information gain,and localizability.Thus,the next observation pose or a series of next poses could be determined with low computational costs.Finally,the exploration of unknown areas could be finished and both the grid map and feature map are constructed accurately.This method reduces the quantity of candidate points,while fuzzy evaluation which imitates human thinking could avoid the effect caused by inaccurate factors.The exploration efficiency is improved.③Multi-steps exploration method using estimation of distribution algorithm is proposed to make use of probabilistic information of grids map.After initializing of robot's tracks from probabilistic information and updating the probabilistic model of their distribution,the optimal path could be efficiently acquired by multiple parallel iterations.The frequency of planning is reduced.④An exploration strategy for mutlti-robots is also established in this dissertation with an improved virtual force method.The speed and exploring direction could be decided by combining the virtual attracting force generated by frontier points,repulsive force generated by obstacles,and the influence caused by distance between the robots.Considering the problem of local minima resulting from the virtual force method,,virtual obstacles and virtual targets are set to help robots to escape such situations and reduce oscillation.Due to the distance control,in this distributed control system,multiple robots could cooperate with each other with few repeating exploration area and collision in movement.Experimental results demonstrate the capability of this approach to escape local minima and improve the efficiency of exploring.⑤Experiments of the proposed exploration algorithms are also conducted on a real robot,the SR-M002 robot.SR-M002 is an omnidirectional mobile robot made by our laboratory.It equips 4 photoelectric encoders and a laser ranger finder.The experiments are completed in the indoor unknown environments.With the results, the effectiveness of above fuzzy evaluation based exploration planning and multi-steps exploration method with estimation of distribution algorithm are verified.
|
PDF Full Text Request