Robot Navigation Environment

A custom OpenAI Gymnasium environment for robot navigation through procedurally generated mazes, featuring both continuous (physics-based) and discrete movement implementations.

Credits: the continuous environment includes heavily modified files originally from OpenAI Gymnasium CarRacing Environment

Installation

To install the required dependencies for the Robot Navigation Environment, follow these steps:

1. Ensure you have Python 3.11 installed on your system.
2. Clone the repository:

   git clone https://github.com/yourusername/robot-navigation-env.git
   cd robot-navigation-env

3. If you have CUDA and GPU acceleration, uncomment the tool.poetry.source and tool.poetry.dependencies sections in the pyproject.toml file:

   # [[tool.poetry.source]]
   # name = "pytorch-gpu"
   # url = "https://download.pytorch.org/whl/cu126"
   # priority = "explicit"
   
   # [tool.poetry.dependencies]
   # torch = {source = "pytorch-gpu"}
   # torchvision = {source = "pytorch-gpu"}
   # torchaudio = {source = "pytorch-gpu"}

4. Install the dependencies using Poetry:

   make build

Features

• Maze Generation: Procedurally generated mazes with configurable parameters
• Robot Dynamics: Physics simulation with realistic movement, including acceleration, braking, and steering
• Pathfinding: A* and Dijkstra's algorithms for optimal path planning
• Multiple Environments:
  • Continuous: Box2D physics with realistic robot dynamics
  • Discrete: Grid-based movement for simpler navigation

Core Components

• Continuous Environment: Physics-based robot navigation

  • Box2D physics engine
  • Customizable robot parameters
  • Real-time visualization with Pygame

• Discrete Environment: Grid-based movement system

  • Simple state representation
  • Fast execution for RL training

• Maze Generation & Pathfinding

  • Configurable maze parameters (size, complexity)
  • A* pathfinding with diagonal movement
  • Path optimization

Usage

# Continuous Environment
from continuous_env.robot_obstacles import RobotObstacles

env = RobotObstacles(render_mode="human")
obs, info = env.reset()

# Action space: [steering, gas, brake, reverse]
action = np.array([0.0, 1.0, 0.0, 0.0])  # Move forward
obs, reward, terminated, truncated, info = env.step(action)

# Discrete Environment
from discrete_env.robot_obstacles_env import RobotObstacleEnv

env = RobotObstacleEnv(render_mode="human")
obs, info = env.reset()

# Action space: Discrete(4) - UP, DOWN, LEFT, RIGHT
action = 0  # Move UP
obs, reward, terminated, truncated, info = env.step(action)

Environment Details

Observation Space

• Continuous: RGB image (96x96x3)
• Discrete: Dict with grid state and agent position

Action Space

• Continuous: Box(-1, 1, shape=(4,))
• Discrete: Discrete(4)

Reward Structure

• Sparse reward on reaching target
• Time penalty to encourage efficiency
• Collision penalties in continuous mode

Dependencies

• gymnasium
• numpy
• pygame
• Box2D (continuous environment only)