Multi-expert learning of adaptive legged locomotion