missing scene folders?

[sha257-lab]

Dear WorldEngine team,

thank you so much for this amazing work!

I stumbled over a little problem and am wondering whether you could help me out: When I run “bash scripts/run_ray_distributed.sh …”, so the command from github.com/OpenDriveLab/WorldEngine/blob/main/docs/simengine_usage.md, run_simulation.py loads scenes_dict from $WORLDENGINE_ROOT/data/sim_engine/scenarios/original/navtrain_50pct_collision/all_scenarios.pkl

scene_dict contains folder names such as (a) 2021.06.09.17.23.18_veh-38_00773_01140-46b1ee12074d5157 or (b) 2021.06.09.17.23.18_veh-38_00773_01140-0dcf9766c1285844 . While folder (a) is present in $WORLDENGINE_ROOT/data/sim_engine/assets/navtrain/assets , directory (b) is not. There are many more missing scenarios.

Is this a mistake on my side or are there folder names in all_scenarios.pkl that are absent from …/navtrain/assets? Do you have any idea what I am missing?

thank you

[WCJ-BERT]

Thank you very much for your detailed investigation and for your interest in WorldEngine. We really appreciate your support and the careful analysis you have done here.

I think the observation you reported is mainly due to the design of MTGS, rather than missing released assets. In MTGS, a reconstructed asset is not necessarily defined for each individual scenario ID. Instead, one asset may correspond to multiple scenario tokens along the same road block / local region. The released asset name is based on a canonical identifier, such as the road block name together with the first central token, while the scenario pkl may contain different scenario IDs or tokens that are still covered by the same reconstructed asset.

Therefore, an exact string match between the scenario ID in the pkl and the asset folder name is not expected to hold in general. The low exact-ID overlap does not necessarily mean that the corresponding reconstructions are missing. The more relevant relation is whether a scenario can be associated with the correct MTGS asset covering the same road block / local region.

For example, 2021.09.14.16.46.51_veh-45_00149_00900-df91d435a2485fca appears in the scenario pkl as a scenario ID, but there is no asset folder with this exact name. This does not mean the asset is missing. The corresponding asset is defined by the MTGS config. In this case, the token df91d435a2485fca is listed under central_tokens in 2021.09.14.16.46.51_veh-45_00149_00900-6bc86e6953cc5004.yaml. So the correct asset ID for this scenario is: 2021.09.14.16.46.51_veh-45_00149_00900-6bc86e6953cc5004.

That said, your report is very helpful because it shows that the current resolving logic or documentation can easily cause confusion, especially when current_scene_id is used as a fallback. We will further clarify the intended scenario-to-asset association mechanism and provide clearer instructions on how to resolve the proper MTGS asset for each scenario.

Thanks again for pointing this out so clearly. This is very useful feedback for improving the usability and documentation of WorldEngine.