After a lot of effort, I finally got an RL trained locomotion policy working in the real robot. It's noticeably smoother than my manually constructed gait. A simple one line bug stymied me for the longest time, but is now fixed!

#robotics #reinfocementlearning

In the world of massive GPU computation resources, what is the role of sample efficiency for RL research? #ReinfocementLearning #RL #Robotics