Guest
This video illustrates how a quadruped learns to jump as high as possible using Bayesian optimization with unknown constraints (BOC). Our results outperform manual tuning. The robot learns to (i) avoid failures and (ii) what is causing them. The former follows from BOC, while the latter is possible with a novel Gaussian process model for the constraint that learns the constraint threshold alongside with the constraint itself. This is a complementary video to our RA-L/ICRA 2021 submission. Take a look to o
