MIT EECS | Aptiv Undergraduate Research and Innovation Scholar
Utilizing Discriminatory Networks and Expert Data in Training Reinforcement Learning Agents
- Artificial Intelligence and Machine Learning
Daniela L. Rus
This project focuses on utilizing uncertainty as an input alongside raw images for learning control in end-to-end autonomous driving. Estimating uncertainty in neural networks remains an open research topic, and actually combining uncertainty with images to create control signals like steering and speed represents a method to increase the safety of neural networks deployed for mobile robotics. This project will necessitate first researching and developing methods for estimating uncertainty, followed by developing a framework for training a model with uncertainty accompanying the typical image input, culminating in testing the model in both new environments and augmentations on the training environment.
“I am excited about SuperUROP for the opportunity to further my research into machine learning for autonomous vehicles. SuperUROP will help expand my research background and prepare me for further research and software work in the future. Through my SuperUROP project, I hope to make autonomous vehicles safer by incorporating uncertainty into the decision making process.”