BridgeData V2 is a large and diverse dataset of robotic manipulation behaviors designed to facilitate research in scalable robot learning. The dataset is compatible with open-vocabulary, multi-task learning methods conditioned on goal images or natural language instructions. Skills learned from the data generalize to novel objects and environments, as well as across institutions.
Dataset Composition
To support broad generalization, we collected data for a wide range of tasks in many environments with variation in objects, camera pose, and workspace positioning. Each trajectory is labeled with a natural langauge instruction corresponding to the task the robot is performing.
- 60,096 trajectories
- 50,365 teleoperated demonstrations
- 9,731 rollouts from a scripted pick-and-place policy
- 24 environments
- 13 skills
Environments
The 24 environments in BridgeData V2 are grouped into 4 categories. The majority of the data comes from 7 distinct toy kitchens, which include some combination of sinks, stoves, and microwaves. The remaining environments come from diverse sources, including various tabletops, standalone toy sinks, a toy laundry machine, and more.
Use case:
Scalable Robot Learning from Diverse Manipulation Behaviors
| Example Observation | RGB image 120 × 160 (camera fixed above workspace) • Depth image 120 × 160 (optional channel) • Robot proprioception: – 7 joint positions, 7 joint velocities – Gripper width (1 value) • Task embedding: 1-of-117 natural-language instruction tokens (“put the blue block into the drawer”, “wipe the spill”, …) |
| Example Action | • End-effector Cartesian displacement Δx, Δy, Δz (3) • End-effector orientation Δroll, Δpitch, Δyaw (3) • Gripper open/close command (1) → 7-dimensional action vector sampled at 10 Hz. |
| Example Reward | • Sparse terminal success flag supplied in metadata (success = 1, failure = 0). • Shaped proxy that can be computed offline: – r_t = (success_t – distance_to_goal_t / d_max) – λ·collision_t. • Business metric: average reward ≅ % successful pick-and-place tasks. |