The BridgeData V2 dataset was jointly released by the University of California, Berkeley, Stanford University, Google DeepMind, and CMU in 2023. The relevant paper results are "BridgeData V2: A Dataset for Robot Learning at Scale".

This is a large and diverse dataset designed to promote scalable robot learning research, containing 60,096 robot trajectories collected in 24 different environments. In order to enhance the generalization ability of robots, researchers collect a large amount of task data in multiple environments with different objects, camera positions, and workspace positioning. Each trajectory is accompanied by natural language instructions corresponding to the robot task. The skills learned from this data can be applied to new objects and environments, and even used across institutions, making this dataset an important resource for researchers.

The characteristics of the BridgeData V2 dataset are:

• Includes a variety of skills: such as pick-and-place, pushing, sweeping, stacking, folding, etc.
• Across multiple environments (24 Environments): The dataset is collected under different environmental conditions, increasing the diversity and generalization ability of the task.
• Open Vocabulary: The dataset supports task specification with open vocabulary, which can be conditioned on learning via target images or natural language instructions.
• Involving more than 100 objects (100+ Objects): The dataset contains operations on a variety of objects, which increases the complexity and practicality of learning.