We present an optimal posture formulation for robotic recovery on uneven terrain and a novel teacher-student reinforcement learning framework that enables a student policy to infer external terrain conditions using only proprioceptive sensing.
We show that our policy performs on the non-flat environments.