Liquid Networks for Learning Control

This lecture by the instructor covers the concept of Liquid Networks for Learning Control, focusing on their application in autonomous systems. The presentation starts with an overview of classical autonomous driving pipelines and the challenges they face. It then delves into the concept of end-to-end learning, emphasizing the importance of raw sensor data in learning control directly. The lecture explores the performance of Continuous-time Neural Networks and their role in modeling physical dynamics. It also discusses the advantages of Liquid Time-Constant Networks in improving representation learning. The session concludes with a discussion on the robustness and performance of Liquid Networks in various scenarios, highlighting their potential for real-world applications.