Collision-free motion planning for dual-arm robots in changing environments

This thesis presents a framework for collision-free motion planning of dual-arm robots in changing environments. The first part deals with global path planning using dynamic roadmaps. In this approach, a preprocessing stage is used to generate a graph in the configuration space of the robot as well as mappings from the discretized workspace to the graph. These mappings enable the fast computation of collision-free paths in the planning stage. In this thesis, five different approaches are proposed and evaluated to plan dual-arm motions with individual dynamic roadmaps for each arm instead of a coupled one. In the second part of the thesis, local planning with an optimization-based approach is considered, whereby the focus is on planning with closed kinematics, which is required if the robot moves an object with both arms together. The task is formulated as an optimization problem and solved numerically with a gradient-based augmented Lagrangian algorithm. The time-critical element of the planner is the computation of the distance to collisions, for which therefore a sphere approximation of the robot and a distance transformation of the environment are employed. In addition, an alternative formulation of the planner as optimal controller with shrinking horizon is proposed. The third part considers reactive planning in changing environments. To this end, three different replanning strategies are investigated. The main contribution is a predictive path-following controller that efficiently moves the robot along the globally planned paths and thereby allows to perform replanning without stopping the robot. Furthermore, the controller can be extended to enable local collision avoidance, which results in a highly robust motion planning framework.