CoolFace
Datasetpublic

jeanmidev/trackmania-community-tracks-and-telemetry

Trackmania Community Tracks & Telemetry Roughly 900,000 community-made Trackmania tracks, collected from the Mania Exchange community track-exchange sites, each paired with its block-level layout and with real human leaderboard-replay telemetry, across two games: Game Tracks TMNF (TrackMania Nations Forever) ~616,800 TM2020 (Trackmania, 2020) ~281,800 Where it comes from Every track and its leaderboard replays were published by their creators to… See the full description on the dataset page: https://huggingface.co/datasets/jeanmidev/trackmania-community-tracks-and-telemetry.

sourceHugging Facecc-by-4.0updated 13d agoView on Hugging Face
1likes777downloads
Dataset Card

Trackmania Community Tracks & Telemetry

Roughly 900,000 community-made [Trackmania](https://www.trackmania.com) tracks, collected from the Mania Exchange community track-exchange sites, each paired with its block-level layout and with real human leaderboard-replay telemetry, across two games:

Where it comes from

Every track and its leaderboard replays were published by their creators to the [Mania Exchange](https://mania-exchange.com) community sites, the long-running, fan-run track-exchange network for the series:

GameSource site
TMNF[TMNF-Exchange](https://tmnf.exchange) (tmnf.exchange)
TM2020[Trackmania Exchange / TMX](https://trackmania.exchange) (trackmania.exchange)

This is community upload data from those exchange sites. It is not scraped from Nadeo's or Ubisoft's own servers. Both games share the same block-based track editor and .Gbx file format, which is what the layouts and telemetry here are extracted from.

What makes it unusual is the pairing. For each track you get both the static design (every block, where it sits on the grid, which way it faces) and the driven reality: the trajectories of many independent human runs through that design, sampled as car position plus the player's steering, throttle, and brake inputs.

What's in it

Per game (tm2020/, tmnf/), four tables. Full column reference in `data_dictionary.md`.

  • —`tracks_details_<game>.parquet`: one row per track. Catalogue metadata (name, author, tags, difficulty, awards, upload date, and so on), the #1 leaderboard time, and per-lap analysis stats (elevation profile, net rotation, speed distribution, corner/straight counts, world-space footprint).
  • —`traces/traces_<game>_partNNNN.parquet`: one row per recorded replay sample, across every leaderboard replay collected, not just the world record (see trace_rank). Each row has car position x/y/z, input_steer/gas/brake, and the nearest placed block.
  • —`blocks/blocks_<game>_partNNNN.parquet`: one row per block placed on the track. Name, grid coordinates, orientation, the order in which the winning run reached it, and replay-derived motion (entry / mid / exit heading and speed, averaged over the runs that actually drove through it, with reliability counts).
  • —`blocks_details_<game>.parquet`: one row per unique block name in that game, with usage counts (n_tracks, n_blocks).

All three per-track tables join on track_id. traces_ / blocks_ are split into partNNNN chunks and a given track_id appears in exactly one part; look up its chunk_part in tracks_details_<game> to open the right part file directly instead of scanning them all.

Block icons

This dataset does not ship block icon images: the icon sets they would come from are community re-uploads with no stated license. The blocks_details_<game> section of `data_dictionary.md` explains how to rebuild a name-to-icon lookup yourself from the original Openplanet packs.

Background

How this dataset was collected and built, a walk through the schema, and what it has been used for so far (procedural level generation, and ideas beyond it), is written up here:

License & attribution

The compilation, schema, and derived analytics columns are released under CC BY 4.0: free to use, including commercially, with attribution.

The underlying material (track layouts, track names, and replay traces) is the work of its original creators and remains subject to Trackmania / Nadeo / Ubisoft terms; it is included here for research and analysis, and its use remains subject to the Mania Exchange sites' own terms. uploader_name and authors are public profile handles from those exchange sites. Contact the dataset owner for correction or removal of specific records.

Citation

bibtex
@misc{trackmania_community_tracks_and_telemetry,
  author       = {jeanmidev},
  title        = {Trackmania Community Tracks \& Telemetry: community-made tracks
                  with block layouts and human replay telemetry (TMNF + TM2020)},
  year         = {2026},
  publisher    = {Hugging Face},
  howpublished = {\url{https://huggingface.co/datasets/jeanmidev/trackmania-community-tracks-and-telemetry}},
  note         = {Licensed CC BY 4.0}
}

Acknowledgements

This dataset is only a wrapper around two decades of other people's work.

  • —Nadeo, for creating Trackmania and building it into one of the richest user-generated-content experiences in racing: an in-game editor anyone can pick up, and the leaderboard and replay system that turns every track into a shared, measurable challenge.
  • —The [Mania Exchange](https://mania-exchange.com) community, and the volunteers who build and run TMNF-Exchange and Trackmania Exchange, for the tracks its players designed, published, and refined, the leaderboard replays that make up the telemetry, and the exchange sites that host and catalogue it all. Every row here is someone's work.
  • —[Donadigo](https://github.com/donadigo), for pygbx, which pioneered open parsing of Trackmania's .Gbx format and informs the extraction approach used here, and for TMTrackNN, an early and influential demonstration that Trackmania tracks can be learned and generated.