| --- |
| license: apache-2.0 |
| task_categories: |
| - robotics |
| tags: |
| - robotics |
| - robotwin |
| - rmbench |
| - chainvla |
| - imitation-learning |
| --- |
| |
| # RMBench ChainVLA 5-Task 500 Diverse Dataset |
|
|
| This dataset packages five RMBench tasks with approximately 500 demonstrations per task for |
| ChainVLA/X-VLA multi-task imitation learning. |
|
|
| Repository: `mmlab-franka/RMBench-ChainVLA-5Task-500-Diverse` |
|
|
| ## Contents |
|
|
| | task | base split | base episodes | supplement split | supplement episodes | total | |
| | --- | --- | ---: | --- | ---: | ---: | |
| | rearrange_blocks | demo_clean | 50 | demo_clean_supp_to500_diverse_20260704 | 450 | 500 | |
| | observe_and_pickup | demo_clean | 50 | demo_clean_supp_to500_diverse_20260704_observe_and_pickup | 450 | 500 | |
| | put_back_block | demo_clean_200 | 200 | demo_clean_supp_to500_diverse_20260704_put_back_block | 300 | 500 | |
| | swap_blocks | demo_clean_200 | 200 | demo_clean_supp_to500_diverse_20260704_swap_blocks, demo_clean_supp_to500_diverse_20260704_swap_blocks_b | 300 | 500 | |
| | swap_T | demo_clean | 50 | demo_clean_supp_to500_diverse_20260704_swap_T_a, demo_clean_supp_to500_diverse_20260704_swap_T_b, demo_clean_supp_to500_diverse_20260704_swap_T_c, demo_clean_supp_to500_diverse_20260704_swap_T_d, demo_clean_supp_to500_diverse_20260704_swap_T_e | 450 | 500 | |
|
|
| Total episodes: `2500`. |
|
|
| ## Layout |
|
|
| ```text |
| data/<task>/<split>/data/episode*.hdf5 |
| data/<task>/<split>/video/episode*.mp4 |
| data/<task>/<split>/instructions/episode*.json |
| data/<task>/<split>/seed.txt |
| data/<task>/<split>/scene_info.json |
| data/<task>/<split>/language_annotation.json |
| metas/<task>.json |
| metas/rmbench500_diverse_20260704_train_meta.json |
| ``` |
|
|
| The temporary `_traj_data` planning cache and local collection logs are intentionally excluded. |
|
|
| ## Training Meta |
|
|
| Use: |
|
|
| ```text |
| metas/rmbench500_diverse_20260704_train_meta.json |
| ``` |
|
|
| Paths in the uploaded meta are relative to the dataset repository root. |
|
|
