| |
|
| [ title ] |
 |
Evolutionary Robot Design & Dynamic Locomotion |
[ abstract ]
 |
|
Goal: a) In a first step, artificial evolution is used to evolve a locomotion strategy for simulated robots with various morphologies by adjusting the weights of their neural network controller. b) A suitable morphology and its corresponding locomotion strategy is chosen and transferred into reality into a physical robot with closed-loop interaction over various sensors and actuators.
Results: The evolutionary approach in the virtual construction kit MorphEngine was investigated on its capacity to inspire and evolve behaviors for legged robots and non-biomimetic locomotion.
The tool supports the engineer in finding, controlling and finally implementing gaits based on emergent dynamics on a real-world robot through the iterative exploitation of both the agents morphology and its physical environment. The iterative, rapid method for digital mock-up and the evolutionary optimisation of a closed loop controller for highly dynamic gaits was used to implement a monkey-type robot. The physical robot was built within 2 days and could successfully exploit the dynamic locomotion strategy previously observed in simulation (see videos below). |
| [ location ] |
|
University of Zurich, Departement of Informatics, Artificial Intelligence Laboratory, Zurich, Switzerland |
| [ framework ] |
|
Semester Thesis I
The project was initiated as a SiROP pilot project. |
| [ duration ] |
|
February - April 2001 |
|
 |
|
|
