Robot Motion Planning and Control
Learning and Optimal Control
Planning real-time motions for robots as complex as humanoid robots in unknown environments remains out of reach to this day. Our work aims to make this goal achievable
- by massively pre-calculating optimal offline movements which are stored in databases,
- by adapting these motions online using predictive control,
- by using data from on-board sensors to control these closed-loop movements.
This work is mainly carried out within the framework of theInstitut National d'Intelligence Artificielle et Naturelle de Toulouse (ANITI) in which Nicolas Mansard holds the Chair of Generation of Movements for complex robots. Olivier Stasse is co-chair . These activities are in line with the European MEMMO project.
Most of the research work relative to model predictive control is integrated into the software platform crocoddyl described in- Carlos Mastalli, Rohan Budhiraja, Wolfgang Merkt, Guilhem Saurel, Bilal Hammoud, Maximilien Naveau, Justin Carpentier, Sethu Vijayakumar and Nicolas Mansard , Crocoddyl: An Efficient and Versatile Framework for Multi-Contact Optimal Control.
Automatic motion planning
The problem consists in automatically computing a collision-free motion for a multi-body system in an environment cluttered with obstacles. This problem can be broken down into various instances depending on whether the moving system is subject to kinematic or dynamic constraints. The latest work by the Gepetto team in this area focuses on manipulation planning. In this instance, objects are moved by robots. This implies specific constraints on the movements of the system: an object not held by a robot must remain motionless, an object held by a gripper is rigidly linked to this gripper.
This work is integrated in the software platform Humanoid Path Planner (HPP) described in
- Prehensile Manipulation Planning: Modelling, Algorithms and Implementation.
The animation below displays a trajectory planned for two robots instructed to assemble two magnetic spheres to a cylinder.
Robot Motion Control
Research on motion control are divided into several themes.
Automatic Synthesis of controllers
This work is linked to the work in manipulation planning which is based on the HPP software platform. From a planned and segmented manipulation trajectory, the principle consists in automatically synthesizing hierarchical task based controllers active on the successive segments of the trajectory. On the trajectory segments where a part of the robot is close to an object (just before a grasp for example), the hierarchical controller contains as a first priority level a task of visually controlling the positioning of the gripper of the robot with respect to the object to be grasped. On the trajectory segments where an object is close to a contact surface (just before putting the object for example), the hierarchical controller contains as a first priority level a task of visually servoing of the relative pose of the object with respect to the contact surface.
This work is integrated into the software platform Agimus.
- Joseph Mirabel, Florent Lamiraux, Long Thuc Ha, Alexis Nicolin, Olivier Stasse and Sébastien Boria Performing manufacturing tasks with a mobile manipulator: from motion planning to sensor based motion control, IEEE International Conference on Automation Science and Engineering, Lyon, France, 2021
The movie below illustrates this work on a deburing task with a mobile manipulator.