We are developing a control design methodology for dynamically walking robots based on recently proven biological control concepts. A direct-search approach involving time/energy minimising test control functions is applied when robot movements are unconstrained. The control learning scheme and parameters to be learnt are similar to those of the human. For stabilising trajectory-constrained motion, we apply robust decentralised controllers designed using optimal trade-off relations between bounds of model uncertainty and control force limits. The efficiency of the proposed control design methodology is verified in dynamics-based simulation of a seven-link biped.
展开▼