Pre-training Contextualized World Models with In-the-wild Videos for Reinforcement Learning

Unsupervised pre-training methods utilizing large and diverse datasets have achieved tremendous success across a range of domains.