Skip to content

Dataset · recast-robotics

Telemetry and video for robot training via DAgger

recast-robotics/rollout_dagger_r1_s2_b0_20260915_121327

Robot trajectory and action recordings at 20 Hz in LeRobot format for imitation learning tasks.

What's inside

The dataset was created within the LeRobot ecosystem (version v3.0) and contains between 1,000 and 10,000 entries. It includes video streams and tabular time series with 8-dimensional action vectors (act_0.pos – act_7.pos) in float32 format, recorded at 20 fps. Judging by the name, these are rollouts collected during training using the DAgger algorithm.

What you can do with it

  • Training robot control policies using imitation learning
  • Fine-tuning models on trajectory corrections within the DAgger loop
  • Testing multimodal robotics pipelines in LeRobot
Tasks
Robotics
Size
1K–10K rows
Downloads this month
21K
Likes
0

How to load it

from datasets import load_dataset

ds = load_dataset("recast-robotics/rollout_dagger_r1_s2_b0_20260915_121327")