Facebook AI Research
JSM - 01 July 2019
From YouTube: https://www.youtube.com/watch?v=g0TaYhjpOfo
Robotics still heavily rely on human expertise !
On one hand, it is unfeasible to hand-design general purpose controllers
On the other hand, there is mistrust for automatic design of controllers
Policy (i.e., parametrized controller)
Learning a controller is equivalent to optimizing the parameters of the controller
Parameters of the policy
Bio-inspired Bipedal Robot "Fox":
[Calandra, R.; Seyfarth, A.; Peters, J. & Deisenroth, M. P. Bayesian Optimization for Learning Gaits under Uncertainty Annals of Mathematics and Artificial Intelligence (AMAI), 2015, 76, 5-23]
Not Symmetrical (about 5° difference). Why?
Because it is walking in a circle!
Question: can we move beyond standard single-objective BO?
[Yang, B.; Wang, G.; Calandra, R.; Contreras, D.; Levine, S. & Pister, K. Learning Flexible and Reusable Locomotion Primitives for a Microrobot IEEE Robotics and Automation Letters (RA-L), 2018, 3, 1904-1911]
[Liao, T.; Wang, G.; Yang, B.; Lee, R.; Pister, K.; Levine, S. & Calandra, R. Data-efficient Learning of Morphology and Controller for a Microrobot IEEE International Conference on Robotics and Automation (ICRA), 2019]
Two levels of optimization
(instead of a single bigger optimization)
Thank you for your time