動画検索
関連広告
検索結果
Before this work, our learned control policies were not robust to stairs.
By training a policy on hundreds of thousands of interactions with stairs in simulation, we can reliably ascend and descend stairs without any perception or knowledge of the environment.
We can achieve sim-to-real transfer to a wide variety of unmodeled, unsensed stairs in the real world.
The control policies are also effective on steep hills and unstructured terrain.