muzilyag/snake_ai

How muzilyag/snake_ai works

This is a multi-agent snake reinforcement-learning application: it runs parallel game environments, trains role-specific MAPPO agents, and offers live visualization, metrics, plotting, and checkpoints. The main loop coordinates environment setup, action selection, simulation, rollout collection, and training. The native engine separates simulation systems for physics, radar observations, and rewards. Source excerpts establish the main Python wiring and some native dependencies; the Python-to-native runtime connection and several unsampled subsystem details are not fully shown, so those links are omitted or marked unverified.

Main entry points: main.py Training loop · vec_env.cpp Vector engine · actor.py MAPPO agent · env_wrapper.py Vector environment

Components 21

Run coordination

Training loop main.py
Settings src/config.py

MAPPO learning

MAPPO agent src/agents/actor.py
Actor-critic models src/agents/mappo_models.py
Rollout buffer src/agents/rollout.py
MAPPO trainer src/agents/mappo_trainer.py

Game simulation

Vector environment src/env/env_wrapper.py
Snake entities src/env/entity.py
Python binding src/engine/pybind_module.cpp
Vector engine src/engine/vec_env.cpp
Physics system src/engine/systems/physics.cpp
Radar system src/engine/systems/radar.cpp
Reward system src/engine/systems/rewards.cpp

Visualization and records

Live renderer src/ui/renderer.py
Game monitor src/utils/monitor.py
Metrics logger src/utils/metrics.py
Metrics plotter src/plotter.py
Checkpoint manager src/utils/checkpointer.py
Reward visualizer src/utils/reward_visualizer.py
Session CSV

Other components

Operator

Connections 25

Operator → Training loop: starts
Training loop → Settings: reads settings main.py
Training loop → Vector environment: creates and steps main.py
Training loop → MAPPO agent: creates and queries main.py
Training loop → MAPPO trainer: dispatches training main.py
Training loop → Live renderer: renders main.py
Training loop → Game monitor: updates main.py
Training loop → Metrics logger: logs metrics main.py
Training loop → Metrics plotter: opens plots main.py
Training loop → Checkpoint manager: saves and loads main.py
Vector environment → Snake entities: uses death reasons src/env/env_wrapper.py
Vector environment → Python binding: uses native extension
Python binding → Vector engine: exposes src/engine/pybind_module.cpp
Vector engine → Physics system: dispatches src/engine/vec_env.cpp
Vector engine → Radar system: dispatches src/engine/vec_env.cpp
Vector engine → Reward system: dispatches src/engine/vec_env.cpp
MAPPO agent → Actor-critic models: uses networks src/agents/actor.py
MAPPO agent → Rollout buffer: creates buffer src/agents/actor.py
MAPPO trainer → MAPPO agent: updates parameters src/agents/mappo_trainer.py
Live renderer → Vector environment: draws world src/ui/renderer.py
Live renderer → Game monitor: reads statistics src/ui/renderer.py
Metrics logger → Game monitor: accepts monitor state src/utils/metrics.py
Metrics logger → Session CSV: writes rows src/utils/metrics.py
Metrics plotter → Session CSV: reads session data src/plotter.py
Metrics plotter → Settings: reads team colors src/plotter.py

More Python diagrams: plumhound/plumhound · shi-labs/versatile-diffusion · nazdridoy/kokoro-tts · open-edge-platform/geti · sense-x/co-detr

Drawn by GitDiagram on September 30, 2026 from the repository's file tree, README and source files. Read as Markdown ·