Motion Planning for Humanoid Robots
A humanoid robot operating in a human-centered environment needs a component to plan collision-free motions in a fast and robust manner. Based on randomized algorithms, such as Rapidly Exploring Random Trees (RRT), we are developing algorithms for collision-free motion planning for single and dual grasping, re-grasping and dual-arm manipulation tasks. In addition, we are investigating grasping pipelines combining several tasks, such as finding a feasible grasp, solving the inverse kinematics and searching the configuration space for collision-free trajectories. The planning algorithms are evaluated and integrated on the humanoid robots of the ARMAR series which are developed by the High Performance Humanoid Technologies lab (H²T), KIT, Germany.
Most of the developed algorithms become part of the open source C++ software framework Simox, which can be downloaded from gitlab. The simulation and motion planning toolbox offers the possibility to simulate complex robot systems (VirtualRobot), to realize grasp planning algorithms (GraspStudio) and to plan collision-free motions (Saba).
Videos
- The Dual-Arm J+-RRT planner is used to re-grasp a plate in a cluttered environment. The results are presented in a real world experiment.
- Manipulation planning with the Dual-Arm IK-RRT planner.
- The Grasp-RRT planner integrates the three main tasks needed for grasping an object: plan a feasible grasp, solve the inverse kinematics and search a collision free motion. The planner is used to solve single arm and bimanual grasping tasks in simulation and on the humanoid robot ARMAR-III.
- The task of finding a cooperative grasping motion for multiple robots is addressed by the Multi-Robot-RRT. All IK-solutions (covering the robot’s base positions), a suitable grasping setup for multiple hands together with a collision free motion of all robots can be computed efficiently. As input the planner needs precomputed reachability information (for efficient IK-solving) and a set of potential grasping positions for each end-effector. The planner is then able to find a multi-robot grasping motion for targets that are defined in workspace (e.g. the Cartesian position of a table). The examples show two exemplary scenes covering up to three humanoid robots resulting in a 60-dimensional planning problem.