Sites Inria

Version française

Scientifique culture

27/10/2010

Small robots capable of big things

© SYMBRION

The five-year SYMBRION project (Symbiotic Evolutionary Robot Organisms) has just passed the halfway mark. Twelve European partners are involved in this collaborative robotics project.

The aim of this cooperation (between robots) is to get small autonomous robots to come together and function as a single autonomous organism. If successful, such a project could have very diverse applications: it could, for instance, be used to explore unknown and dangerous terrains or to find earthquake survivors. Robots could, for instance, form a ‘snake’ to pass through a narrow corridor, then split up to search a wider area, or transform themselves into a ‘spider’ to get over obstacles, etc.

 The robots are programmed by evolution. It is important to note that only the programs controlling the robots will actually ‘reproduce’. The robots themselves will not multiply! However, the best-performing programs for the accomplishment of a given task will spread from robot to robot, meaning that the robots effectively end up sharing a part of their brains.

Some experiments of this kind have already been performed, but only ‘offline’, meaning that the evolution phase takes place under supervision, and that only the best programs resulting from the evolution are then installed on real robots and used to control them during the execution of the task.
The scientific challenge of this project is to perform the same procedure online. The robots are not shut down and must be able to improve through evolution without external help: this brings us very much into the territory of artificial intelligence.

This artificial intelligence project will eventually enable us to explore dangerous terrains and find earthquake survivors

The Tao project-team is working on two aspects of this collaborative project:

  • The definition of evolution criteria. The programs must be evaluated so that the best ones can be propagated, but on what criteria should this evaluation be made? The idea is to try to give robots "instincts" such as curiosity and a spirit of discovery.
  • How can this Darwinian evolution be instilled in robots? Existing artificial evolution algorithms cannot be used in their present form, and must be adapted to each particular environment.

The difficulty will then be to manage the relationship between these two points. Having established the evolution criteria, it is important for these criteria to be optimised in order to take into account the new characteristics of the best programs upon each new implementation. The robots are ultimately designed to perform a specific task, and the aim is to see how evolution can be guided without restricting the field of possibilities by imposing too specific an idea of how the task in question should be performed.

© SYMBRION This image illustrates the artificial intelligence concept developed by the SYMBRION project, but is not representative of real experimentation.

The Tao project-team is therefore working on an intangible part of this robotics project, while other partners define and build the robots themselves. But designing and building new types of robots takes time, and while we wait for these robots to become available, we must work in the abstract, with no possibility of testing the results on real robots, which is obviously the ultimate test in robotics.

The complexity of the problem stems from the fact that the algorithms have to work on real robots and not only on the simulators used during their design. This ‘reality gap’, which stems from the fact that a simulator does not take account of all of the hazards of real life – such as the possibility of a robot’s wheels slipping – raises the question that has troubled robotics experts for decades: how can you develop robotics without a robot? The challenge, though, is a fascinating one.

Keywords: Inria Saclay - Île-de-France Tao Artificial Intelligence Marc Schoenauer Robots

Top