Doctorant sous la direction de P. Rodriguez

Titre de la thèse: Adaptive strategies for collision detection and control of lightweight robot manipulators – application to safe human-robot interaction
Résumé de la thèse: This PhD has started on October 1st, 2015 and is carried out in collaboration with the Interactive Robotics Laboratory (LRI) of the French Alternative Energies and Atomic Energy Commission (CEA).

Context
In the few past years, a growing interest of the research community and industry is focused on the development of lightweight robots and the associated control laws for safe human-robot interaction, with many potential applications in industry, aerospace, medical or service robotics. In manufacturing, for the tasks in the production line which are difficult to automate, robot manipulators with collaboration capabilities can help making repetitive tasks less painful for the operator. When a human user and a collaborative robot share the same workspace, the robot manipulator must ensure a safe behaviour through its mechanical design but also with reactive control strategies. The robot controller must be able to detect collision at the earliest and then respond appropriately and securely to limit damage due to physical interaction with the operator.

Objectives
The objective of this research project is to develop an effective strategy for collision detection between a robot manipulator and its environment, and the associated post impact control laws, in the presence of modelling uncertainties and using a minimal number of sensors (without force/torque sensors).

Methodology
In order to make possible the comparison, both theoretical and experimental, of the performance of existing detection strategies, a quantitative methodology for the analysis of collision detection algorithms will be developed. In continuation of previous work at the Automatic Department of CentraleSupélec and with the CEA-LIST, an adaptive strategy of filtering and decision taking into account the state of the system will be studied through the synthesis of observers for the detection, taking into account the influence of the robot's transmission flexibilities on detection. Once the collision is detected, an impedance control law with gain adaptation will be considered to ensure an efficient behavior in post-impact phase relatively to performance and robustness objectives (passivity, transparency) and taking into account the specific characteristics of the detected impact (contact point, magnitude of effort) and task.

The proposed methods will be experimentally evaluated using a lightweight robot arm developed at CEA LIST. This six degree of freedom robot, mechanically backdrivable and equipped with motor and joint position sensors, represents a privileged test bench for the evaluation of collision detection methods and control for safe human robot interaction.