Sovereign 8s.is mirror of standardgalactic/GTQN
  • Python 99%
  • Dockerfile 0.5%
  • Shell 0.5%
Find a file
Repository files (latest commit first)
Filename Latest commit message Latest commit date
2026-02-24 20:22:14 +05:00
docs paper_mapping.md 2026-02-24 20:22:14 +05:00
gtqn Delete gtqn/configs/ablations/empty 2026-02-24 01:06:22 +05:00
GTQN_sumo_assets_corridors HOWTO_BUILD_NET.txt 2026-02-24 20:18:45 +05:00
scripts Delete scripts/empty 2026-02-24 00:58:58 +05:00
Dockerfile Add files via upload 2026-02-24 00:37:08 +05:00
LICENSE Add files via upload 2026-02-24 00:37:08 +05:00
pyproject.toml Add files via upload 2026-02-24 00:37:08 +05:00
README.md README.md 2026-02-24 20:17:01 +05:00
requirements.txt Add files via upload 2026-02-24 00:37:08 +05:00

GTQN - Graph Transformer Q-Network

Graph Transformer Q-Network for Collaborative Governance and Decentralized Decision-Making in Multi-Intersection Networks

It includes:

  • Decentralized Distributed Junction Controller (DJC) with two-stage sparse coordination (discrete peer gating + soft relevance weighting)
  • Centralized Training / Decentralized Execution (CTDE) via a Collaborative Governance Graph (CGG)
  • Unified spatiotemporal representation: temporal Transformer + graph-aware coordination
  • SUMO/TraCI environment wrapper for multi-intersection control (event-driven, phase-duration actuation)
  • Reproducible configs and ablation toggles (attention, reward, governance)

Quickstart

1) Install SUMO

Ensure sumo and sumo-gui are in PATH and SUMO_HOME is set.

2) Python env

python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

3) Prepare a scenario

Put a SUMO scenario under:

gtqn/envs/networks/<scenario_name>/
  - net.net.xml
  - sim.sumocfg
  - routes.rou.xml   (or generate using scripts/make_routes.py)
  - tls_ids.json     (list of controlled traffic light IDs)
  - corridor.json    (optional: corridor lanes + entry junction)

4) Train

python scripts/train.py --config gtqn/configs/default.yaml   env.scenario=gtqn/envs/networks/SQ1   train.total_env_steps=1200000

5) Evaluate

python scripts/eval.py --run_dir runs/<your_run>

License

MIT (see LICENSE).