Abstract
This paper presents a general fuzzy reinforcement learning (FRL) method for biped dynamic balance control. Based on a neuro-fuzzy network architecture, different kinds of expert knowledge and measurement-based information can be incorporated into the FRL agent to initialise its action network, critic network and/or evaluation feedback module so as to accelerate its learning. The proposed FRL agent is constructed and verified using the simulation model of a physical biped robot. The simulation analysis shows that by incorporation of the human intuitive balancing knowledge and walking evaluation knowledge, the FRL agent's learning rate for side-to-side and front-to-back balance of the simulated biped can be improved. We also demonstrate that it is possible for a biped robot to start its walking with a priori knowledge and then learn to improve its behaviour with the FRL agents.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call