Yuzhe Qin
Jun Wang
Kaiming Kuang
Yigit Korkmaz
Description: DESCRIPTION META TAG
keywords should be placed here (63)
--> By incorporating extensive data augmentation in a simulated environment, CyberDemo outperforms traditional in-domain real-world demonstrations when transferred to the real world, handling diverse physical and visual conditions. Regardless of its affordability and convenience in data collection, CyberDemo outperforms baseline methods in terms of success rates across various tasks and exhibits generalizability with previously unseen objects. For example, it can rotate novel tetra-valve and penta-valve, de
CyberDemo Pipeline. First, we collect both simulated and real demonstrations via vision-based teleoperation. Following this, we train the policy on simulated data, incorporating the proposed data augmentation techniques. During training, we apply automatic curriculum learning, which incrementally enhances the randomness scale based on task performance. Finally, the policy is fine-tuned with a few real demos before being deployed to the real world.
Auto Curriculum Learning. We begin training from the simplest level, L = 0, signifying no augmentation, and then evaluate the task success rate after several steps of training. The evaluation difficulty aligns with the current level of L. When the success rate surpasses a pre-defined threshold, we advance to the next level, which brings greater augmentation and harder evaluation. If the success rate fails to reach the threshold, we create additional augmented training data and stays at the current level. We
Yuzhe Qin
Jun Wang
Kaiming Kuang
Yigit Korkmaz