Go to content Go to the menu Go to the search

ResiBots: Robots with Animal-like Resistance - Jean-Baptiste Mouret

Quick access, personalized services


Advanced search


Direction de la communication


Marie Pinhas-Diena, in charge of scientific communications l tel: +33 (0)1 44 27 22 89 l email: marie.pinhas@upmc.fr

ResiBots: Robots with Animal-like Resistance, Jean-Baptiste Mouret

Despite over 50 years of research in robotics, most existing robots are not nearly as resilient as animals: they are fragile machines that can easily quit working when the going gets tough. The project goal is to change that by offering new algorithms for autonomous robots to continue their missions in totally unexpected situations, including material damage.


Jean-Baptiste Mouret, Research Director


The current approach to fault tolerance comes from the approach used for many critical systems such as spacecraft or nuclear power plants. It is difficult to implement on more “mainstream” autonomous robots because it relies on automatic diagnostics, which require many internal sensors and thus make them expensive and complex robots. The choice of contingency plans depends on the quality of diagnosis and cannot respond to all situations.


In this project, the opposite of a classical approach is taken by eliminating the concept of diagnosis. It is not necessary to understand the problem in order to find a way to continue the mission. Instead of a diagnostic approach/process, the project uses learning algorithms by trial and error (reinforcement learning): The robot does some tests to find functions that work despite the damage. Our goal is to make algorithms that enable a robot to find a new behavior in less than twenty tries and less than 2 minutes, i.e. much faster than what can be done now. The project tests this idea using three types of robots: a mobile robot with an arm (designed by the project team), a hybrid hexapod robot with wheel lugs (designed in the team at UPMC) and the humanoid robot iCub (designed at IIT, in Italy).


The hexapod robot is used in the experiments to evaluate our algorithms. Here it has a broken left front leg. The robot loads a battery, a computer with GNU / Linux, and a 3D camera. With a visual odometry algorithm, the 3D camera enables the robot to evaluate its movement speed and therefore assess its performance without an external device. © Antoine Cully / UPMC 2015


At the end of the project, the team plans to put a robot in a room, break certain mechanical parts (e.g. cutting a flap in two or lock the arm), and observe the robot learn in minutes to independently compensate for damage. For example, the walking robot could learn to walk with only 5 feet in all directions, and taking into account the obstacles for learning.

Expected results

The purpose of this project is to propose and validate new algorithms to adapt to robot damage. In general, no one wants to buy a robot that becomes defective with any small problem: it is often preferable that the robot can continue to operate as long as possible, even though its performance is degraded. Imagine, for example, a rescue robot used after an earthquake to look for survivors. If the robot is partially broken, for example because of falling debris, it should be able to continue its mission without requiring rescuers to take time to come and help.


Ultimately, most autonomous robots could be equipped with these algorithms, such as service robots (e.g. assistance robots to help people, which wear out and deteriorate), industrial robots (intervening there where it is difficult to send a human), rescue robots (which help rescuers following catastrophes) or even robots that are beginning to be used in agriculture.

For more information:

This project will take place at Inria - Nancy Grand Est, in the Larsen team (Long-term Autonomy and interaction skills for Robots in a Sensing ENvironment). Before the start of this project, Jean-Baptiste Mouret was at The Institute of Robotics and Intelligent Systems (Isir, UPMC/CNRS/Inserm).Nouvelle fenêtreNouvelle fenêtre


La page personnelle de Jean-Baptiste Mouret.Nouvelle fenêtre (in English)


Le portrait de Jean-Baptiste Mouret sur le site de l'INS2I du CNRS.Nouvelle fenêtre (in French)