Low-level
Low-level policy is trained to perform goal-reaching in simulation. A number of randomizations are applied during training, including randomized height fields (terrain).
Low-level policy is trained to perform goal-reaching in simulation. A number of randomizations are applied during training, including randomized height fields (terrain).
Low-level behaviors generalize to real-world environments.
Low-level behaviors generalize to real-world environments.
Appropriate randomizations are necessary during training.
Appropriate randomizations are necessary during training.
No randomizations in training.
All randomizations except height field.