This has been a mystery to me.
All the walking robots look like idiots now. But we do have a lot of simulation-based results (Flexible Muscle-Based Locomotion for Bipedal Creatures ), so why can't we just apply the simulation results to a real robot and let it walk, not like an idiot, but like an running ostrich?
With the main loop running at more than 60 fps, I fail to see how possibly the program could fail to stop the robot from losing balance. When I balance a stick on my hand, I could probably only do 5 fps.
We have not only supercomputers connected to the robots, but also reinforcement learning algorithms at our disposal, so what has been the bottleneck in the problem of bipedal walking?