Conq Hose Manipulation: 139 Robotic Episodes and 8,277 Frames

Conq Hose Manipulation contains 139 robotic episodes and 8,277 frames of hose manipulation data across three distinct tasks. Developed by lerobot using the LeRobot library, the data includes synchronized video and time-series telemetry captured at 30 FPS. The dataset is structured into Parquet files for state data and MP4 files for visual records.

Use Cases

• Training imitation learning models using the 8,277 frames of video and state data
• Temporal modeling of deformable object manipulation using the 30 FPS time-series data
• Evaluating multi-task robotic performance across the 3 defined task categories

Strengths

• 139 distinct robotic episodes
• 8,277 frames of synchronized video and state data
• High-frequency 30 FPS capture rate
• Apache 2.0 permissive license

Limitations

• Small sample size of 139 episodes
• Robot hardware type is not specified in the metadata
• Limited task variety with only 3 tasks represented

Provenance

Source

lerobot

Collection Method

Sensor and video capture from robotic hardware via the LeRobot library

Freshness

Last updated March 2026.

Geography

United States