Hugging Face's logo

luxury-lakehouse
/
pitch-control

Other
English
numpy
sports-analytics
soccer
football
tracking-data
pitch-control
physics-model
heuristic
Model card Files
xet
 Community

Pitch Control — Physics-Based Team-Control Probability Surface

Computes a per-frame probability surface indicating which team controls each location on the pitch. Physics-based heuristic following Spearman (2017). No trained weights — the method is an independent Python translation of the published equations.

Part of the (Right! Luxury!) Lakehouse soccer analytics platform. Pitch Control is the load-bearing substrate for Off-Ball xT, Space Creation, OBSO, and PAUSA.

Method Description

Pitch Control estimates each player's influence over every grid cell on the pitch using a time-to-intercept (TTI) function parameterised by player position and velocity. The probability that a team controls a cell is the sum of its players' influence values divided by the total influence across both teams, producing a pair of complementary probabilities at every cell (home-team and away-team, summing to 1).

Algorithm

For each tracking frame:

  1. For each player, compute time-to-intercept for every target cell given their current position, velocity vector, and a maximum-acceleration constraint.
  2. Convert TTI to a logistic influence function: lower TTI → higher influence; the logistic slope is a model parameter.
  3. Sum influence per team; normalise across teams to produce a probability bounded in [0, 1] per cell per team.
  4. Output: a 2-D probability surface per frame, one grid per team.

Reference

Inputs

No training data — this is a heuristic, not a learned model.

Runtime inputs per frame:

Field Source Description
Player positions (x, y) Tracking vendor (Metrica / IDSSE / SkillCorner) 105×68 m canonical pitch, metric units
Player velocities (vx, vy) Tracking vendor Differenced positions smoothed over a short window
Team attribution Tracking metadata home / away per player
Grid resolution Configuration Default: 68×52 cells (~1.5 m resolution)

The canonical tracking corpus used by the project is published as luxury-lakehouse/pitch-control-tracking on HuggingFace Hub and attributed in NOTICE.

Execution

Computed daily via Databricks serverless workflow compute_pitch_control (module ingestion.pitch_control_batch). Distribution: applyInPandas grouped by synthetic frame_batch_id partitions to stay within the 1 GB UDF memory limit. Output table: {catalog}.bronze.pitch_control_values.

A batched Numba-accelerated implementation is available for off-line surface generation. Benchmark target: ≤ 5 ms per frame for 22 targets.

See workflow-cards/wf-pitch-control.yaml for the full operational contract.

Intended Use

  • Substrate for downstream analytics: Off-Ball xT, Space Creation, OBSO, PAUSA
  • Tactical visualisation: Team-control regions at key moments on the dashboard's Pitch Control page
  • Research: Reproducible implementation of Spearman's model on open tracking data

EU AI Act — Intended Use and Non-Use

This method is published for research and reproducibility purposes on public, open-licensed tracking data. It is not intended for, not validated for, and not supplied to any use that would fall within Annex III §4 (Employment, workers management and access to self-employment) of Regulation (EU) 2024/1689 — including recruitment or selection of natural persons, decisions affecting work-related contractual relationships, promotion, termination, task allocation based on individual traits, or the monitoring and evaluation of performance and behaviour of workers for employment decisions.

Any deployer who wishes to use this method for such a purpose is responsible for performing their own conformity assessment under Article 43, for drawing up the technical documentation required by Article 11 and Annex IV, for implementing the human oversight measures required by Article 14, for declaring accuracy metrics under Article 15, and for ensuring the data governance obligations of Article 10 are met. Note specifically that this is an unparameterised physics heuristic with no training data — so representativity, bias, and protected-attribute audits are not directly applicable to the method itself, but the downstream player-level metrics derived from it (Off-Ball xT, Space Creation, OBSO, PAUSA) inherit the representativity of the tracking corpus on which they are evaluated.

See the AI_GOVERNANCE.md gap analysis in the source repository for the project's full risk classification, re-classification triggers, and governance posture.

Limitations

  • Heuristic, not learned. Parameters (TTI logistic slope, maximum acceleration) are set by published values and have not been fit to any dataset. The method does not adapt to playing style, pitch conditions, or data-source characteristics.
  • No defender intent model. All players react identically. Defenders with off-ball tracking or engagement intent are not distinguished.
  • Tracking-data quality dependence. Missing players, skipped frames, or velocity-smoothing artefacts propagate directly into the surface.
  • 2-D only. Aerial play (headers, clearances) is ignored.
  • No ball physics. Ball trajectory is not explicitly modelled; the surface describes player control, not ball control.

Files

No model weights. The method is implemented in source:

  • src/analytics/pitch_control.py
  • src/ingestion/pitch_control_batch.py

Outputs are published as the Delta table {catalog}.bronze.pitch_control_values and in Parquet form as luxury-lakehouse/pitch-control-tracking.

Citation 

@inproceedings{spearman2017physicsbased,
  title={Physics-Based Modeling of Pass Probabilities in Soccer},
  author={Spearman, William},
  booktitle={MIT Sloan Sports Analytics Conference},
  year={2017}
}

@software{nielsen2026pitchcontrol,
  title={Pitch Control: Physics-Based Team-Control Probability Surface on Open Tracking Data},
  author={Nielsen, Karsten Skytt},
  year={2026},
  url={https://github.com/karsten-s-nielsen/luxury-lakehouse}
}

Companion Resources

Resource Description
Pitch Control Tracking Canonical tracking corpus formatted for pitch control computation
OBSO Trained Grids Downstream EPV transition and reachability grids

More Information

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
Other
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Dataset used to train luxury-lakehouse/pitch-control