How dvidnguyen/mapf-benchmark-system works
This is an in-progress multi-agent pathfinding (MAPF) benchmark for comparing A*, tabular Q-learning, and DQN on grid maps. The main workflow configures a training run, steps agents through the collision-aware environment, updates learning agents, and logs episode metrics; evaluation runs trained or baseline agents and records benchmark results. Separate visual tools support algorithm comparison and step-by-step A* playback. The graph follows the sampled source and README; unsampled map/scenario loading and some view wiring are not expanded beyond documented or indexed relationships.
Main entry points: evaluate.py Evaluation CLI · main.py Training CLI · grid_env.py Grid world · compare_ui.py Algorithm comparison UI
Components 13
Run orchestration
- Training CLI
main.py - Evaluation CLI
evaluate.py - Episode runner
core/simulation.py - Benchmark metrics
evaluate.py
MAPF environment
- Grid world
envs/grid_env.py
Planning and learning
- Agent contract
agents/base_agent.py - A* planner
agents/a_star.py - Q-learning
agents/q_learning.py - DQN agent
agents/dqn.py
Views and comparison
- Algorithm comparison UI
UI/compare_ui.py - Grid canvas viewer
gui/canvas_viewer.py - A* visual runner
run_astar_visual.py
Other components
- Benchmark user
Connections 20
- Benchmark user → Training CLI: starts training
README.md - Benchmark user → Evaluation CLI: runs evaluation
evaluate.py - Training CLI → Grid world: configures world
main.py - Training CLI → Q-learning: creates agents
main.py - Training CLI → DQN agent: creates agents
main.py - Training CLI → Episode runner: runs episodes
main.py - Episode runner → Grid world: resets and steps
core/simulation.py - Episode runner → Agent contract: calls agent methods
core/simulation.py - Evaluation CLI → Grid world: configures world
evaluate.py - Evaluation CLI → A* planner: creates planner
evaluate.py - Evaluation CLI → Q-learning: creates agent
evaluate.py - Evaluation CLI → DQN agent: creates agent
evaluate.py - Evaluation CLI → Episode runner: runs test episodes
evaluate.py - Evaluation CLI → Benchmark metrics: writes results
evaluate.py - Algorithm comparison UI → Grid world: steps world
UI/compare_ui.py - Algorithm comparison UI → A* planner: compares planner
UI/compare_ui.py - Algorithm comparison UI → Q-learning: compares agent
UI/compare_ui.py - Algorithm comparison UI → DQN agent: compares agent
UI/compare_ui.py - A* visual runner → Grid world: steps world
run_astar_visual.py - A* visual runner → A* planner: uses planner
run_astar_visual.py