TransportEnergyGridEnv

TransportEnergyGridEnv (deathenv)

A custom Gymnasium environment for Reinforcement Learning. The agent operates in a grid environment to pick up cargo, manage energy reserves, avoid moving/patrolling enemies, and safely deliver cargo to the destination.

---

Features

• Gymnasium-compliant: Inherits directly from gymnasium.Env and integrates seamlessly with gym.make().
• Rich State Space: Includes dynamic local grid channels (Obstacles, Enemies, Energy Barrels, Goal/Cargo) and status meta-features (Energy level, Step progress, Target direction).
• Customizable Grid: Adjustable grid size, obstacle counts, energy barrel frequency, and dynamic enemy patrols.
• Built-in Video Recorder: Simple .record() method to evaluate policies and save MP4 demonstrations.

---

Installation

From Hugging Face

pip install git+https://huggingface.co/minzSleeps/TransportEnergyGridEnv.git

From GitHub

pip install git+https://github.com/Minz-sleeps/DeathEnv.git

From source

git clone https://github.com/Minz-sleeps/DeathEnv.git
cd DeathEnv
pip install -e .

---

Quickstart Usage

import gymnasium as gym
import deathenv

env = gym.make("TransportEnergyGrid-v0", grid_size=20, n_obstacles=10)

obs, info = env.reset(seed=42)
done = False

while not done:
    action = env.action_space.sample()
    obs, reward, terminated, truncated, info = env.step(action)
    done = terminated or truncated

env.close()

Recording a Demo Video

from src.env import TransportEnergyGridEnv

env = TransportEnergyGridEnv(grid_size=20)
env.record("demo.mp4", model=None, seed=42)
env.close()

---

Repository Structure

DeathEnv/
├── pyproject.toml          # Package configuration and dependencies
├── README.md               # Documentation
├── requirements.txt        # Core dependencies
├── src/
│   ├── __init__.py         # Package exports & Gymnasium registration
│   └── env.py              # TransportEnergyGridEnv implementation
└── examples/
    ├── train.py            # Example: Train DQN agent via Stable-Baselines3
    ├── evaluate.py         # Example: Evaluate model & record video

---

Environment Details

Action Space

• 0: Move Up
• 1: Move Down
• 2: Move Left
• 3: Move Right

Observation Space

Local grid surrounding the agent (6 channels) + 5 metadata attributes (Normalized energy, Has cargo flag, Step progress, Normalized X/Y distance to target).

Rewards Configuration

The environment features fully customizable rewards via the rewards argument (a dictionary overriding default values):

custom_rewards = {
    "step": -0.05,               # Step penalty
    "collision": -2.0,           # Obstacle or grid boundary collision
    "cargo_pickup": 30.0,        # Cargo pickup reward
    "delivery": 100.0,           # Successful delivery to destination
    "barrel": 20.0,              # Collecting energy barrel
    "enemy_collision": -50.0,    # Collision with an enemy
    "energy_exhaustion": -30.0   # Running out of energy
}


env = gym.make("TransportEnergyGrid-v0", rewards={"delivery": 200.0, "step": -0.1})

---

Running Examples

Training a DQN agent using [Stable-Baselines3]:

python examples/train.py --timesteps 1000000 --grid-size 30

Evaluating a trained model and saving a video:

python examples/evaluate.py --model-path dqn_transport_agent --output demo.mp4

---