How nitin3560/uav-marl-task-allocation works
This repository is an experimental multi-UAV task-allocation and relay-routing study. Its main workflow trains or selects a joint policy/baseline, runs it against the task-and-communication environment, and evaluates task completion, delivery, energy, throughput, and AoI; analysis plots compare logged outcomes. The environment’s task, network, energy, and reward models are distinct internal stages. Evaluation relationships are supported by the eval-runner reference list and excerpts; some training and plotting connections are documented only at a high level, so those are marked unverified.
Main entry points: eval_runner.py Episode evaluation · lawn_task_comm_env.py Parallel environment · training Policy training · marl_policy.py MAPPO policy
Components 16
Swarm environment
- Parallel environment
envs/lawn_task_comm_env.py - Task lifecycle
envs/task_models.py - Network routing
envs/network_models.py - Energy accounting
envs/energy_models.py - Team reward
envs/reward.py - Motion helper
controllers/motion_controller.py
Policies and training
- Policy training
training - MAPPO policy
agents/marl_policy.py - CTDE wrapper
utils/ctde.py - Baseline policies
baselines
Evaluation
- Episode evaluation
evaluation/eval_runner.py - Fault injection
evaluation/fault_injector.py - Evaluation metrics
metrics
Results analysis
- Result plots
analysis/plots.py - Evaluation logs
Other components
- Experimenter
Connections 16
- Experimenter → Policy training: trains policies
- Policy training → Parallel environment: trains on
- Policy training → MAPPO policy: optimizes
- CTDE wrapper → Parallel environment: wraps
utils/ctde.py - Episode evaluation → Parallel environment: steps episodes
evaluation/eval_runner.py - Episode evaluation → MAPPO policy: loads and calls
evaluation/eval_runner.py - Episode evaluation → Baseline policies: selects and calls
evaluation/eval_runner.py - Episode evaluation → Motion helper: calls
evaluation/eval_runner.py - Episode evaluation → Fault injection: configures
evaluation/eval_runner.py - Episode evaluation → Evaluation metrics: computes
evaluation/eval_runner.py - Parallel environment → Task lifecycle: uses
envs/lawn_task_comm_env.py - Parallel environment → Network routing: uses
envs/lawn_task_comm_env.py - Parallel environment → Energy accounting: uses
envs/lawn_task_comm_env.py - Parallel environment → Team reward: uses
envs/lawn_task_comm_env.py - Result plots → Evaluation logs: reads logs
analysis/plots.py - Episode evaluation → Evaluation logs: writes results