Explore more domain randomization methods

[manx52]

randomize the frictionof the ground,
add noise to the observations
randomly push the robots during the episode to teach them a more stable stance.
Each robot has a friction coefficient sampled uniformly in [0.5,
1.25].
The pushes happen every 10 s.
The robots’ base is accelerated up to ±1 m/s in both x and y directions.