"Café des Sciences", March 21st 2011, Ludovic Righetti

Café des Sciences du 21 Mars 2011
Speaker: Ludovic Righetti, PhD
Title: Are personal robots going to simplify our lives soon?
Subject: robotics, computer sciences

2011’s second « Café des Sciences » took place at the University of Southern California (USC) on March 21st and featured a presentation by Dr. Ludovic Righetti.

Ludovic Righetti is a postdoctoral researcher at the Computational Learning and Motor Control Lab (CLMCL) at the University of Southern California since March 2009. He studied at the Ecole Polytechnique Fédérale de Lausanne where he received an engineering degree in 2004 and a Doctorate in Science in 2008. His doctoral thesis was awarded the 2010 Georges Giralt PhD Award given by the European Robotics Research Network for the best robotics thesis in Europe. His research focuses of the development of control methods for agile robots able to move in difficult environments and able to manipulate objects.

He first described the main tasks personal robots would complete in the future. Robots with defined shapes (human-like or not) would be able to carry out very diverse missions. They could for example help elderly and/or disabled people in their daily lives or take care of household chores. They could be used in difficult or hazardous environment and serve as rescue agents after disasters. Finally, progress in robotics could make the creation of smart prosthesis a reality and offer replacements that precisely mimic the function of the lost limb. The field of possible applications is very large and the coming “robotics revolution” could create a society where personal robots are as casual a sight as cell phones nowadays.

Yet additional developments are required to unlock these possibilities. Event though robots are already working in our factories, they are not autonomous and are unable to perform other tasks than those they have been programmed to carry. Even though they are extremely accurate in their action, they are unable to adapt to the slightest change in their environment. Research currently carried at the CLMCL aims at developing the autonomy and flexibility of robots by making them able to move, handle unknown objects, learn new skills, perceive and analyze their environment and interact safely with human beings. To do so, researchers study the motor system and learning process of living animals, as they are perfect examples of autonomous robots.

Ludovic Righetti described several projects that are being developed in his laboratory and aimed at developing learning processes into different kinds of robots. Two learning methods are used at the CLMCL : imitation learning and reinforcement learning. In the former case, a human operator “shows” the robot how to perform an action in ideal conditions (pick up an object for example) and the learning process focuses on learning how to perform this action in a non-ideal environment (the object is moving or an obstacle hinders the robot for example). In the latter case, the robot is programmed to learn to complete a task through trial and error: after each attempt, it modifies its movement pattern, until it is able to complete the task. Using this method, it is possible for a dexterous robotic arm to learn how to pick up an object in no more than a hundred trials.

The CLMCL applies this research to locomotion in rough environments with the Little Dog robot. This quadruped robot is trained to move on a rough surface, that it perceives through cameras. The locomotive algorithm of the robot computes around a hundred visual parameters, which allows it to autonomously analyze its environment and pick the best path to get through it. The robot learns to detect good footholds by trial and error: each time it makes a step, a human controller rates the foothold as good or bad. After training, the control algorithm is able to autonomously rate a potential footfall through analysis of its visual characteristics. Hence, like in the case of living animals, experience is the key to learning for the robot. The CLMCL also works on the flexibility of locomotive algorithms, by introducing the notion of inverse dynamics control, which allows the robot to adapt its course and its move when confronted to unexpected obstacles.

Several key points for the further development of autonomous robots were discussed in the Q&A session following the presentation. The link between hardware (the robot) and software (the programs and algorithms that control it) is still a critical aspect of research in this domain and it is nowadays hardly possible to break it : programs must be written for a specific robot and robots must be constructed bearing in mind what kind of program will be used to control them. Data processing and response time are also among the current chokepoints for the development of autonomous robots: due to the large amount of data that needs to be processed by environment analysis and decision-making programs, autonomous robots are not nearly as fast as pre-programmed robots. On the bight side, international cooperation in this domain is well-developed: data and result exchanges are numerous and facilitated by the fact that many algorithms (such as those developed at USC) are open-source. This is a very positive and encouraging characteristic of this field and such a research environment should foster breakthroughs in the future.

Click here for movies and additional information about the research currently carried at the CLMCL

Last modified on 28/03/2011

top of the page