Evolving Robust Control Strategies for Simulated Animats
31 Mar 2009
Recent research in the field of robotics has seen an increase of interest in the topic of modular, self-reconfiguring robotic systems. Along with recent improvements to automatic fabrication methods and rapid prototyping technologies, these advances have given researchers the ability to experiment with different robot morphologies quickly and with little cost.
The problem of high-level, task-oriented control of these robots however, has remained a challenge. The reasons for this are threefold. First, researchers face a huge space of possible robot morphologies from which they can choose from. Second, these robot configurations tend to have many more degrees of freedom than traditional, hand-built robots. Third, since many of these robots bare no resemblance to creatures we are familiar with in nature, designing control strategies for them by hand can often be difficult.
This project introduces a framework which identifies the morphologies which are most suited to the tasks the robot is expected to perform. In addition, the framework uses evolutionary optimisation techniques to automatically find motor control strategies for any given robot morphology.
Downloads
Videos
Snake Locomotion Stream Download |
Snake Rotation Stream Download |
|
Starfish Locomotion Stream Download |
Starfish Rotation Stream Download |
|
Insect Locomotion Stream Download |
Insect Rotation Stream Download |
|
The Simulation Environment Stream Download |
Fun stuff
Work in progress videos: Getting the physics simulator working for the first time. Improving the renderer. More than just boxes. Experimenting with joints and rag-dolls. Trying out the engine on a manually-controlled biped. Hooking up the physics engine with the GA routines. Creating realistic morphologies.