The incremental evolution of gaits for hexapod robots

摘要

Gait control programs for hexapod robots are learned by incremental evolution. The first increment is used to learn the activations required to generate a single leg cycle. At this level the control program is required to produce the proper sequence of pulses needed to generate smooth movement by the servos. The learning program needs to take into account the peculiarities of the servo, its mounting and the capabilities of the leg. The second increment of the learning process is used to learn the best combination of individual leg cycles to produce a gait. This part requires the learning system to choose the best leg cycles for each leg and to coordinate their movement. In this paper, we describe an application of this method to learn gaits for an actual hexapod robot. A cyclic genetic algorithm is used to learn efficient gait cycles for each leg. A genetic algorithm is used to combine these leg cycles in such a way that coordinated gaits result. Tests are conducted on the actual robot to confirm the method's viability.