You are hereMotivational scenario for embodied evolution

Motivational scenario for embodied evolution


Our motivational scenario for studying embodied evolution envisions an environment and a group of (simulated) robots that has to adapt to the environment, or perform a user defined task, or both. The principal problem here is that the link between the robot makeup (hardware and software) and robot behaviour is ill-understood, noisy, and depends strongly on the environment that may change over time. Using evolution to mitigate this problem is a natural idea and in general there are two approaches for doing this:

  • evolving the software (the controllers) for robots with a fixed hardware (morphology), or
  • evolving the software (the controllers) and the hardware (morphology) together.

Furthermore, we can distinguish off-line evolution trying to optimize the design before the robots are launched and on-line evolution continually adapting the design during the operational period.

The greatest challenge is clearly the on-line evolution of both the controllers and the morphologies. Such an evolutionary system requires self-reproducing robot organisms, i.e. a system where robot "babies" can be produced with inherited software and hardware attributes. Our group is doing fundamental research in this area developing the conceptual framework, the algorithms, and proof of concept implementations. For obvious reasons much of this work is in simulation. Possible research questions include issues about reproduction mechanisms, the combination of environmental and performance-based fitness measures, or the combination of lifetime learning and evolution.

Example publications of CI group members and students:

Background and more information:

  • Futuristic paper on grand challenges for evolutionary robotics: here.
  • Recent state of the art overview: here.
  • The classic book about evolutionary robotics: here.
  • A more recent book on evolutionary robotics: here.