RTE-UPMC Research Chair

Identifiez-vous pour accéder à la partie privée du site ou le modifier. Déconnexion.

Research Topics

The research activity focuses on three major aspects:

1. Intrinsically safe control of intervention robots evolving under constraints and in interaction with human operators

The working context of the applications of this chair is  such that: the trajectories associated to the considered tasks are mostly not known in advance, it induces the evolution of the robot in a potentially cluttered and dynamic environment and a human operator can potentially be engaged in a close-range interactions with the robot. Given these constraints, the activities of the chair in the domain of control aim at developing control laws dedicated to robotic manipulators involved in both autonomous and comanipulation tasks  and which ensures:

• the safety of the human operator. This implies guaranteed collision avoidance in the cases where autonomous work is envisioned with a human operator in the range of motion of the robot. In comanipulation tasks, contact is unavoidable but induced forces and torques have to remain within some acceptable range thus implying the control of the impedance of this unavoidable physical interaction.
• the proper coexistence of several, potentially conflicting, tasks and constraints for the robot. This implies a formulation of the control problem that allows to hierarchically combine and accommodate these tasks and constraints in a generic fashion.
• the proper coexistence of several autonomous and collaborative modes: load bearing, gravity compensation, force amplification, assistance to precise manipulation, repetitive tasks...

2. Automatic {Constraints and Tasks}-based design of physical architecture/morphology of intervention, comanipulated robots
Comanipulation allows to keep the human expertise in the loop. Several questions are raised by this new paradigm and it is important to account for the human presence at the very first stage of design. A potential solution which we explore within the context of this chair rely on an automatic approach. It is based on multi-objectives optimization to propose potential morphologies for the optimal achievement of some tasks in some environments. The concept is as follows: a population of robots is automatically generated that spans a large range of potential solutions, the performances of these robots are evaluated in simulation scenarii through quantitative indicators. Some optimization policy is applied which yields a new population of robots. This cycle is applied until some convergence criteria is met. When human operators are not involved, the performance indicators can be rather straightforward: task tracking errors, complexity of the robot, force capabilities... However, when humans come into play, one needs to find indicators which can quantify, in an automatic and repeatable fashion, the quality of the physical human-robot interaction in the particular case of an industrial task where the operational performance remains the main objective even though often contradictory with the comfort of the operator. Our work in this domain thus aims at: "discovering" the right indicators to use and formulating the optimization problem in order to benefit from state-of-the-art multi-objecives optimization techniques.

3. Validation through realistic physics simulation

Both at the control and design level, there is a need for quantitative evaluation of the operational performance induced by the physical and control architecture of the robot as well for quantitative evaluation of the safety of robot interactions with its environment. In particular, one of the key point lies in our capability of generating complex virtual mannequin behaviours in order to simulate human operators. As a consequence, a large effort is put within the framework of the chair on: developing a software architecture dedicated to both realistic physics simulation and experimental validation and pursuing our on-going effort in virtual character multi-objectives control.