Skip to content

Latest commit

 

History

History
26 lines (21 loc) · 1.3 KB

File metadata and controls

26 lines (21 loc) · 1.3 KB

src/

Python RL pipeline for training and running an agent in Geometry Dash via the Geode shared memory bridge.

Packages

Package Purpose
gdrl/env/ Gym environment, Geode IPC adapter, live monitor, shared memory utilities — see gdrl/env/README.md
gdrl/teacher/ Teacher PPO training using privileged game observations
gdrl/data/ Rollout collector — records teacher episodes to NPZ shards
gdrl/student/ Student distillation trainer — learns to mimic teacher from frame data
gdrl/infer/ Inference stub for running a trained student agent
gdrl/common/ Shared config

Key files

File Purpose
gdrl/env/geode_ipc.py Shared memory adapter that reads game state from /gdrl_ipc
gdrl/env/privileged_env.py Gym env wrapping the IPC adapter with reward logic
gdrl/env/live_monitor.py Prints live telemetry from a running GD session
gdrl/env/geode_wait.py Waits for the shared memory segment to appear (connectivity check)
gdrl/teacher/train_ppo.py Trains teacher PPO agent, saves to artifacts/
gdrl/data/collector_cli.py Runs teacher rollouts and saves NPZ datasets
gdrl/student/train_distill_dataset.py Trains student from NPZ via knowledge distillation