Deep RL - Virtual Robotic Arm

Project Overview

This project creates a simulation for a robotic arm that uses deep reinforcement learning. The goal is to train a neural network to control the arm to reach a certain target point in the environment, which in this case is an apple.

Implements a 2-DOF (degrees of freedom) robotic arm simulation using simple math and physics to create realistic arm movements with proper joint limits.

The real-time Pygame visualization provides complete feedback on both the physical system state and the learning progress. The arm is rendered with scaled components including a base platform, rotating joint connections, segments, and a gripper.

64 Neural Network Neurons

DQN Learning Algorithm

2D Physics Simulation

Real-time Visual Feedback

AI Learning Demonstration

Watch the neural network learn to control the robotic arm through reinforcement learning

Real Life Applications

Diverse applications demonstrating the power of AI-driven robotic control

Research In Reinforcement Learning

It shows how effective RL is for controlling tasks and the challenges it faces, leading to more research in AI training.

Education

It can be used as a teaching tool for learning about robotics, AI, and control systems.

Game Development

Players can interact with a virtual world to build robotic mechanisms making experimentation with robotics principles possible.

Simulation Training

Controlled environment for testing algorithms before deploying them on actual robotic systems.

Project Gallery

Creating individual robot components and a reward object using GIMP

Base

Arm

Joint

Gripper

Reward

Technical Specifications

Detailed breakdown of the AI system and learning algorithms

Core Components

Deep Neural Network: DQN with PyTorch to control the arm's movements, using 64 neurons in the network.
Reinforcement Learning: Memory queue to store previous experiences for training the model.
Visual Feedback: Visual feedback of the arm's position and the target location, improving user interaction.
Reward System: Rewarding efficient movements and penalizing collisions.
Custom Images: Ability to load custom images for the robotic arm, gripper, joint and target.

Technology Used

PyTorch DQN Pygame NumPy Neural Networks Matplotlib