Merging Global and Local Planners: Real-Time Replanning Algorithm of Redundant Robots Within a Task-Priority Framework

Abstract

Task-priority inverse kinematics is a popular motion control algorithm which efficiently handles redundancy in robot manipulators. It has been recently extended in order to handle also set-based control objectives or inequality constraints. As any local motion planner it is prone to the occurrence of local minima. This work further extends set-based inverse kinematics by adding a motion planner in order to avoid such occurrence. Motion planners are usually computationally heavy especially in their eventual implementation with a task-priority architecture. To reduce this issue, the planner is implemented as a sampling-based algorithm which works in the reduced-dimensionality of the robot workspace applying Cartesian constraints only. The output trajectory is then checked against the inverse kinematics algorithm exploiting the redundancy and verifying the fulfillment of the joint-based task constraints. During the motion, inverse kinematics is then used also in real-time to ensure a reactive behavior to address, e.g., mismatch between the a-priori information and real-time perception acquisition. Also, the motion planner runs in background to adapt to changes in the environment or to accommodate incremental mapping. Comparison with alternative approaches are investigated and discussed. The most promising method is validated first in hundreds of numerical simulations to provide a solid statistical analysis and then experimentally with a Kinova Jaco2 7 DOFs manipulator equipped with an RGB-D sensor. Note to Practitioners—In this work we propose a motion planning algorithm that allows to effectively deploy robot manipulators in partially unstructured environments. It is structured in two layers: a Cartesian space global planner guarantees the feasibility of the trajectory with respect to potential obstacles in the environment, while a reactive local planner checks the feasibility at joint level. This architecture allows to define all the safety constraints both at Cartesian and joint space needed to operate in unstructured environments, while keeping the computation time lower with respect to standard approaches that define all the constraints in an offline global planner. The reduced computation time allows to effectively perform real-time replanning operations when needed, making the robotic system capable of adapting to a dynamic environment and suitable for sharing its workspace with human operators.