This work was done as part of a research project in the Neural Networks Research Group at UT Austin (http://nn.cs.utexas.edu/) under the mentorship of Dr. Risto Miikkulainen and Dr. Cem Tutum.
The goal of this project was to develop a neural network architecture with improved generalization capabilities, and specifically to test the generalization capabilities in the CARLA environment. Three neural network architectures are trained to drive a vehicle on a set of tracks under a set of varied conditions, and their generalization capabilites are evaluated by analyzing the performance of these vehicles on a set of unseen tracks and unseen conditions.
Instructions:
Training
Evaluation
Analysis