Hybrid planning for free climbing robots: combining task and motion planning with dynamics and control

Official URL: https://risc01.sabanciuniv.edu/record=b2586035_(Table of contents)

Robots with articulated limbs that can free-climb vertical surfaces have the potential to be instrumental in a wide range of applications, ranging from search-and-rescue to surveillance, inspection/maintenance to planetary exploration. Free climbing is highly challenging as it requires the robot to make progress using only friction at the contact points, without using any special equipment. To free-climb vertical terrain, the robot must go through a continuous sequence of configurations satisfying certain constraints that ensure that the robot is in equilibrium, collision-free, and can be controlled within the actuator torque limits. A feasible free-climb plan requires i) deciding on a proper sequence of holds to reach, ii) finding collision-free trajectories for the relevant arms of the robot reach these holds, while utilizing the internal degrees of freedom of the robot to maintain friction contacts and balance of the robot, and iii) ensuring that these trajectories can be executed within the actuator torque limits. Therefore, geometric reasoning and motion planning alone are not sufficient to solve these problems, as the planning of reach actions need to be integrated with the motion planning and the feasibility of plan executions in terms of maintaining friction contacts, balance and actuation capabilities needs to be verified. We propose a hybrid planning approach for free climbing robots that combines high-level representation and reasoning with low-level geometric reasoning, motion planning, balance, and actuator feasibility checks. The hybrid planning approach features bilateral interaction between high-level reasoning and feasibility checks. The high-level reasoner guides the motion planner by finding an optimal task-plan; if there is no feasible kinematic/dynamic/controls solution for that task-plan, then the feasibility checks guide the high-level reasoner by modifying the planning problem with new constraints. We present a validation of our approach through a comprehensive set of benchmark instances and a systematic evaluation its performance in terms of scalability, solution quality, and success rate.