avbiswas/navigation-mappo-rl

How avbiswas/navigation-mappo-rl works

Multi Agent Navigation trains and evaluates cooperative agents in configurable obstacle courses. The main workflow is to choose a YAML scenario, train MAPPO against the multi-agent environment, save a model, then run inference in the original or another scenario. The environment supplies observations and rewards; its geometry, sensing, and rendering are distinct supporting stages. The README and source index establish the principal links, but several source excerpts are partial, so unsampled behavior is not expanded into additional flows.

Main entry points: environment.py Multi-agent environment · mappo.py MAPPO trainer · train_mappo.py Training entry point · inference.py Inference runner

Components 16

Scenario and world

Scenario YAMLs configs
Environment schema nav/config_models.py

Simulation and sensing

Multi-agent environment nav/environment.py
Obstacles and boundary nav/obstacles.py
Geometry utilities nav/utils.py
Ray intersections nav/ray_intersection.py

MAPPO learning

MAPPO trainer rl/mappo.py
Observation encoder networks/actor_critic_network.py
Gaussian actions rl/distributions.py
Rollout buffer rl/rollout_buffer_with_states.py

Rendering

Render snapshot models nav/renderer_models.py
Live renderer nav/live_renderer.py

Train and evaluate

Training entry point train_mappo.py
Inference runner inference.py
Saved model

Other components

Experimenter

Connections 29

Experimenter → Scenario YAMLs: selects README.md
Scenario YAMLs → Environment schema: follows schema README.md
Experimenter → Training entry point: starts training README.md
Training entry point → Scenario YAMLs: loads config README.md
Training entry point → Multi-agent environment: trains in train_mappo.py
Training entry point → MAPPO trainer: configures train_mappo.py
Training entry point → Observation encoder: configures train_mappo.py
Multi-agent environment → Environment schema: uses config nav/environment.py
Multi-agent environment → Obstacles and boundary: uses geometry nav/environment.py
Multi-agent environment → Geometry utilities: uses utilities nav/environment.py
Multi-agent environment → Ray intersections: computes sensing nav/environment.py
Multi-agent environment → Render snapshot models: creates snapshots nav/environment.py
Multi-agent environment → Live renderer: updates display nav/environment.py
Obstacles and boundary → Geometry utilities: uses utilities nav/obstacles.py
Ray intersections → Obstacles and boundary: intersects shapes nav/ray_intersection.py
Ray intersections → Environment schema: uses geometry types nav/ray_intersection.py
Render snapshot models → Environment schema: uses config types nav/renderer_models.py
Render snapshot models → Ray intersections: uses ray results nav/renderer_models.py
Live renderer → Render snapshot models: renders snapshots nav/live_renderer.py
MAPPO trainer → Observation encoder: encodes observations rl/mappo.py
MAPPO trainer → Gaussian actions: uses action distribution rl/mappo.py
MAPPO trainer → Rollout buffer: stores rollouts rl/mappo.py
MAPPO trainer → Multi-agent environment: collects experience rl/mappo.py
MAPPO trainer → Saved model: saves checkpoint README.md
Experimenter → Inference runner: starts evaluation README.md
Inference runner → Multi-agent environment: steps simulation inference.py
Inference runner → MAPPO trainer: loads and predicts inference.py
Inference runner → Scenario YAMLs: selects config README.md
Saved model → Inference runner: loads checkpoint README.md

More Python diagrams: alpha-unito/streamflow · andrestorres123/delve · automateyournetwork/mcpyats · betagouv/comparia · byexamples/byexample

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