Bayesian-optimization study tunes walking controllers on ATRIAS hardware

Evidence notes

• The paper applies Bayesian optimization with a domain-informed feature transform to bipedal locomotion controller tuning.
• The method is evaluated on the ATRIAS biped in simulation and on hardware, including two walking controllers on the real robot.
• It gives ATRIAS a concrete learning-on-hardware research record for sample-efficient bipedal gait optimization.