Cooperative Path Planning for Autonomous Ground Vehicles using 3D Sensor in Cluttered Environment

摘要

This thesis is concerned with the real-time path planning of multiple autonomousground vehicles (AGVs) in a cluttered environment. In order to perform real-timeoperations with limited processing resources, an efficient path-planning algorithm,identification of the obstacles by a single sensor and cooperative path planning aredeveloped.AGVs are widely utilised for scientific, commercial, industrial and military applicationsfor different tasks such as exploration in hazardous environments and unknownarea, military surveillance and reconnaissance, search and rescue missions and industrialautomation. For an AGV, path planning in a cluttered environment is achallenging task due to its lack of information about the surroundings and its needto re-plan its path quickly whenever it senses obstacles nearby. Therefore, an efficientpath-planning algorithm that offers an AGV sufficient time to re-plan its pathto avoid moving obstacles is proposed and, to measure its computational efficacy,its time complexity is considered.Initially, the Efficient D* Lite algorithm, a modified D* Lite graph search algorithmusing the Fibonacci heap data structure to speed up the re-planning process in alarge dynamic environment, is developed and the Pioneer 3DX (P3DX) mobile robotis used as the AGV platform for experiments.The dynamic equations of motion of robot which rely mainly on physical characteristics,such as the robot and actuator specifications, are derived using Lagrangianformulation to model an actual P3DX robot’s behaviour by taking into accountthe vehicle’s kinematic and dynamic constraints. This simulated model is used tovalidate the path-planning algorithm before implementing in real time experiments.In real-time experimentation of autonomous path-planning, AGV relies completelyon perception system to sense the immediate environment and avoid obstacles whenit traverses towards the goal. As the Time-of-Flight (ToF)-based PMD (PhotonicMixer Device) three dimensional (3D) sensor can provide range and intensity dataat low computational cost, it is utilised as a single proprioceptive sensor to detectstatic and dynamic obstacles. The bistatic model and calibration of the PMD camera system are described to analyse the process of image acquisition.As the future motions of moving obstacles are a priori unknown in dynamic environments,it is essential to estimate them based on observations of the obstacles pastand present states so that the AGVs path can be re-planned in advance to avoidcollision in critical conditions. An approach which combines the differential sceneflow technique and the gradient vector field (GVF) for estimating the kinematicbehaviours of the obstacles using the range intensity value of the PMD camera isproposed.To ease the complexity and spatial sparseness in single AGV system in exploringlarger area, an effective approach for cooperative multi-AGV system is proposed. Asmulti-robot systems can be easier, cheaper, more flexible and more fault-tolerant,and cover more space than using one expensive powerful robot, cooperative multi-AGV path planning is addressed. The hybrid and distributed control architecturefor a multi-AGV approach is highlighted in which each AGV has its own intelligenceto re-plan its path and shares its kinematic and sensor information globally withother AGVs.The PMD camera is mounted on a P3DX and integrated with its software library.The Efficient D* Lite algorithm and scene flow technique are successfully implementedon the P3DX onboard system to carry out real-time autonomous tasks.Finally, real-time experiments are performed with three AGVs in different scenariosin order to demonstrate all the path-planning features and validate the proposedapproaches. Videos of the experiment are presented in a CD-ROM (Appendix D).