Bayesian optimization is a fairly generic tool for global optimization of black-box functions that are extremely expensive to evaluate, i.e., where experiments are very costly. The idea behind Bayesian optimization is to exploit previous experiments to build a model/approximator of the unknown objective function and to optimize this model to determine the next experiment.
Bayesian Optimization for Bipedal Locomotion
A key challenge of bipedal locomotion is finding gait parameters, which optimize a desired performance metric, e.g., robustness or energy efficiency. Typically, gait optimization requires extensive robot experiments or specific expert knowledge.
We applied Bayesian optimization to automate and speed up the process of gait optimization to a biologically inspired bipedal walker to maximize the walking speed and reliability.
Roberto Calandra, TU Darmstadt
Nakul Gopalan, TU Darmstadt
Andre Seyfarth, TU Darmstadt
Jan Peters, TU Darmstadt