mulligan2026MIT
Thread Nut: round 5 evaluation
Real-robot held-out evaluation dataset for the Mulligan paper containing episodes from evaluation sessions with multiple policy comparisons (HG-DAgger, HG-DAgger+Mulligan, HiL-IDQL+Mulligan) on thread nut tasks.
Downloads991
Episodes50
Why This Matters for Physical AI
This dataset provides real-world evaluation benchmarks for imitation learning and learning-from-human-feedback methods on robotic manipulation tasks, enabling assessment of policy performance and comparison across different training approaches.
Technical Profile
- Modalities
- rgb
- Environment
- lab
- Task Types
- manipulation
- Episodes
- 50
- Data Format
- parquet
- Annotation Types
- language_instructionsreward_labelsaction_labels
- License
- MIT
Access
Need custom rgb data?
Claru builds purpose-built datasets for lab applications with dense human annotations and quality assurance.
Request a Sample Pack