Omdena-Quantum-Self-Driving Project 2023

Project Lead: Raghav Thiruvallur (Please contact [email protected] for any questions)

Variational Quantum Convolutional Neural Network for Image Recognition to Determine Steering Angle for Autonomous Vehicles

Goal

Design and implement a Variational Quantum Algorithm alongside a classical Machine Learning model to determine the steering angle correction for self-driving vehicles

Motivation

• Existing classical deep neural networks yield good performance
• Aim for more generalized models and higher efficiency while training and testing
• Can we harness the power of quantum computing to help increase generality and/or efficiency?

Project Approach

1. Dataset:
   • Find a dataset which contains front-camera view of the road and corresponding steering wheel angle change to keep the car straight.
   • Understand the dimensions and variety of input images
   • Clean and preprocess the dataset
2. Build the Quantum and Classical Models
   • Group 1: Create a classical Convolutional Neural Network (CNN) to predict the vehicle's steering angle based on images
   • Group 2: Simultaneously build and test a Variational Quantum Algorithm (VQA)
     • Use a simple existing vanilla neural network in place of the CNN
     • Build various ansatze circuits and compare their performances (e.g., convergence, accuracy, loss, runtime, etc.)
3. After initial development, combine CNN and VQA, and fine-tune the model
4. Import the model and work with vehicle-environment simulations (e.g., CARLA, AirSim) and test hybrid model in the environment.

Project Architecture

Deep Learning Neural Networks + Variational Quantum Circuit

Training models using 4 qubit HVA circuit + Deep Learnning Neural netwwrk