Simulation And Navigation Technologies For Smart Wheelchairs In Complex Environment

Mobility limitation has become a primary factor in human functionallimitations. Wheelchair, including electric and smart wheelchair, is the mostintelligent way to improve the independent mobility of the target population.Nowadays, development of more autonomous and secrecy smart wheelchair ishot topic at home and abroad.How to safely perform the driver training on them, and then identifywhether the user has the necessary skills to drive a wheelchair, and how toreduce the cost and shorten the development cycle of smart wheelchairexperiments are the current difficulties encountered in the practical engineering.Starting from this point, we designed and implemented an open simulationplatform for smart wheelchairs. Through analysis of the system structure, alayer structure of the platform including model layer, abstraction layer and theinterface layer is proposed. This platform is implemented based on ROS andGAZEBO, and can be applied to driver training and algorithm designing. Sinceno powered wheelchair training and evaluation standards is currently available,we devised one according to the demands of the driving in daily life. Thisstandard can be implemented both on real and simulated platform, which in aresults, constitutes a training system. It is believed that the driving skills can beimproved and whether the user is able to drive powered wheelchair can bedetermined through this training system.When come to the navigation of wheelchair, the definition of navigationindicators is defined, including Obedience, Safety and Comfort. To optimizethe control commands according to these indicators, a common hierarchicalcontrol combining local planning and global planning is applied. Two parts ofthis control system including map building and local path planning is the key to make wheelchair suitable in complex environments. So a sampling basedlocal path planning method and a multilayer matching based incrementalsimultaneous localization and mapping (SLAM) method are proposed. Thesampling based local path planning method takes obedience, safety, andcomfort as the objective function, and performs a sampling process is in thevelocity space, which is followed by a kinematics simulation and evaluation tofind the optimal control. Since the geometric and nonholonomic constraints istaken into consideration, not only the task specified by the global planner canbe achieved, but also the safety and comfort of the user is ensured. As a datacorrelation problem followed by a least squares optimization, the process ofSLAM is simplified in the context of the multilayer matching basedincremental SLAM. The data association is solved by a multilayer ICP matchin which the fisher criterion describes its uncertainty. An incremental QRdecomposition is applied to solve the graph optimization problem effectively.Avoiding the local minimum, this method can build a consistent map of theenvironment. The real-time application both indoor and outdoor can beachieved due to the low complexity and high robustness of the algorithm.Finally, the experiments verify the effectiveness of the proposedsimulation platform and algorithms.