Grid World Path Planning using PPO

A simple experimental project using Proximal Policy Optimization (PPO) from OpenAI's Spinning Up library, applied to a custom Grid World environment for path planning.

This is an active work-in-progress (WIP). Currently experimenting with:

• 🧭 Increasing action space (more directional controls)
• 🎮 Integrating imitation learning for guided policy initialization
• ⚙️ Exploring environment variations

Overview

The goal is to train an agent to navigate a 2D grid world, reach the target efficiently, and avoid obstacles using reinforcement learning.

Current Setup:

• Environment: Custom Grid World
• Algorithm: PPO from OpenAI Spinning Up
• Experiments:
  • Action space scaling
  • Imitation learning integration
  • Custom reward shaping

Quick Start

1. Clone the repo, install dependencies

2. Run training:

   python algorithms/ppo/ppo.py

Work In Progress

• PPO baseline training
• Expand action space
• Add imitation learning
• Experiment with multiple targets

Acknowledgements

• OpenAI Spinning Up
• PPO Algorithm: "Proximal Policy Optimization Algorithms"

---