Warehouse Testbed for Reinforcement Learning

This repository provides a configurable warehouse simulation environment and reproducible testbed for benchmarking reinforcement learning (RL) algorithms.

The environment models:

• stochastic item arrivals
• dynamic order generation
• constrained agent movement
• action masking (invalid actions filtered)

Built on top of this environment, the testbed enables:

• reproducible scenario generation across heterogeneous warehouse layouts
• evaluation under diverse operating conditions (from low-load to highly congested settings)
• dataset export (CSV) for offline analysis
• training and benchmarking with PPO / MaskablePPO

---

Installation

git clone <your-repo-url>
cd warehouse-testbed
pip install -r requirements.txt

---

Quick Start

Run a simple random agent:

python run_example.py

This will:

• initialize the environment
• render the warehouse
• sample valid actions
• print rewards

---

Environment

Main file:

environment/env_storehouse.py

Gym-style usage:

state, _ = env.reset()
state, reward, terminated, truncated, info = env.step(action)

Action Masking

mask = env.get_available_actions()

---

Scenario Generation

Enable with:

record_scenario=True

This generates:

• *_config.csv → layout and parameters
• *_steps.csv → initial state + events

Useful for:

• reproducibility
• dataset generation
• benchmarking

---

Benchmark Pipeline

Run:

python benchmark_pipeline.py

This will:

• generate scenarios
• train RL agents
• evaluate performance
• save results in results_*

---

Configuration

Defined in:

environment/conf.json

You can modify layouts, probabilities and warehouse settings.

---

Structure

warehouse-testbed/
├── README.md
├── requirements.txt
├── benchmark_pipeline.py
├── run_example.py
├── environment/
│   ├── env_storehouse.py
│   └── conf.json

---

Use Cases

• RL benchmarking
• warehouse optimization
• action masking research
• dataset generation

---

License

MIT License