The researchers assembled an impressive dataset for pretraining, comprising 52 datasets with over 200,000 robot trajectories across four categories. This approach allows robots to learn from a wealth of experiences, including human demonstrations and simulations.
You are viewing a single comment's thread from: