Evolution of Walking Gaits

Designing or optimizing control systems for legged locomotion is a complex and time consuming process. Human engineers can only produce and evaluate a limited number of configurations, although there may be numerous competing designs that should be investigated. Automation of the controller design process allows the evaluation of thousands of competing designs, without requiring prior knowledge of the robot’s walking mechanisms [Ledger 1999]. Development of an automated approach requires the implementation of a control system, a test platform, and an adaptive method for automated design of the controller. Thus, the implemented control system must be capable of expressing control signals that can sufficiently describe the desired walking pattern. Furthermore, the selected control system should be simple to integrate with the adaptive method.

One possible method for automated controller design is to utilize a spline controller and evolve its control parameters with a genetic algorithm [Boeing, Bräunl 2002], [Boeing, Bräunl 2003]. To decrease the evolution time and remove the risk of damaging robot hardware during the evolution, a dynamic mechanical simulation system can be employed.