Study On The Task-Oriented Path Planning Methods For An Omni-Directional Mobile Dual-Arm Robot

The Omni-directional Mobile Dual Arm Robot (OMDAR) has advantages for its agileand wide space moving ability and the ability to imitate the operation of the human’sdual-arm. The combined structure of the mobile platform and dual arm improves not only thedexority and adaptability, but also the difficulty to planning and control. The missiondefinition in configuration space of the robot is very complicated, because the robot’s cmplexstructure and high degrees of freedom. Therefore, mission definition in Descartes space ormore abstracted definition in higher level can simplify the robot application and missiondescription. For support the simplified mission definition, realization the map from themission space to robot’s configuration space is vital, and must carefully design the pathplanning algorithm to bring into play the ability of the robot.This thesis focuses on realizing the methods for mapping from mission space toconfiguration space, and takes the SmartPal service robot, designed by Yaskawa ElectronicsCompany, as the object and the prototype of the OMDAR system. The effort is mainlydevoted to study the planning methods for two kinds of mission, which includes missionsdefined by the continuous track or discret pose configuration constraints. The continuoustrack mission needs the planning method to realize on-line resolution of the reversekinematics of the OMDAR, and the mission defined by discrete pose configurationconstraints needs the planning method to plan the transition movement between the discreteconfiguration constraints for the OMDAR.For better understanding and analysis of the robot’s kinematics and dynamicscharacteristics, this thesis analyzed and modeled the kinematics and dynamics of theOMDAR system as left invariant form with screw theory and Lie group notations. Theanalyzed expression of acceleration screw is setuped for the tree topology multi-rigid-body system. The traditional Lie bracket operation is extended to a generalized form. Based on theJourdain variant principle and the reciprocal product of the wrench and twist, a legible andcanonical dynamic model representing the relation between the inputs and the generalizeddynamic load wrenches is presented. The screw form Kane dynamics equation can be got inthe modeling process. Using the proposed modeling method, the dynamics model of the robothas clear geometrical and succinct block factorizations form. This makes the interactionanalysis of different part of the robot system convenient.The OMDAR system has components with similar function, such as the left and rightarm. And the system has same requirements for the each link’s movement, such as obstacleavoidance. On the other hand, the path planning algorithm design must cope with differenttypes, dimensions and range of a group of optimization criteria simultaneously for optimizingthe system’s movement. Based on above demands and the characteristics of the robot and itsmission, the product arbitration multi criteria optimization with constraints is setuped as thebasic model to resolve the path planning problem of the OMDAR. For simplifying thevariable search, the quotient exterior point penalty method is proposed to convert theconstrainted optimization problem to unconstrainted one. The product criteria has complexloacal extrenum point structure, traditional local gradient based optimization methods areprone to convergent to the local extrenum point. Therefore, a Gaussian Rovering ParticalSwarm Optimization (GR-PSO) method is proposed to search the global extrenum point. TheGR-PSO method absorbed the advantage of the traditional simplex method and the randomevolution Partical Swarm Optimization (PSO) method.For designing the planning method to realize the continueous track mission, this thesisproposed new obstacle avoidance, singularity avoidance, tip-over avoidance and stabilityoptimization methods based on the robot’s characteristics and the production arbitration multicriteria optimization model. The methods are verified accordance with the simulation results.The obstacle avoidance mechanism is designed based on the critical point’s velocityprojection on the vector that points from the link’s critical point to the nearest point of theobstacle. The singularity avoidance method is not only suitable for the control of redundantmanipulators but can be also applied for non-redundant manipulators. When applied to redundant manipulators, the method alleviates the burden of checking the nature of thesingularity points. The generalized dynamics of mobile manipulators is analyzed. For a planarsupporting region, the tip-over avoiding requirement is formulated as the tip-over preventconstraints with the reciprocal products of the resultant support wrench and the imaginarytip-over twists, which are constructed with the boundaries of the convex support polygon. Anew on-line stablility optimization method for mobile robot is proposed based on thecloseness of the support polygon’s boundaries.For designing the planning alogorithm to realize the mission defined with two discretepose or position configuration constraints. First, this thesis designs Local Muli-level Extendand Connect Random Rapid-exploring Tree (LMEC-RRT) algorithm based on RandomRapid-exploring Tree (RRT) algorithm. Simulation experiments manifest that LMEC-RRThas high efficiency and high success rate attributes. Then the LMEC-RRT algorithm isextended to cope with the mission with closed chain constraints. Application of picking andplacing a box for the OMDAR is utilized to verify the viability of the alogorithm. The RRTmethod or its variants have shortcoming in planed path somoothability, because the plannedpath is constructed by connectting a group disrete viable configurations with short linesections. To overcome the shortcoming, this thesis designed a Sliding Window OptimizationMethod (SWQM) to construct the time based and smooth path in configuration space. TheSWQM is setuped based on the product arbitration multi criteria optimization model and theparametered curve model which is used to connect the discrete configurations. Based on theGR-PSO solver, the SWQM is verifed to be successful in time based and smooth pathoptimization by simulation experiments.