Explore the intersection of reinforcement learning and mixed autonomy traffic systems in this 32-minute lecture by Cathy Wu from MIT. Delve into the challenges of quantifying the impact of technology on societal systems, focusing on transportation in the US from 2020 to 2049. Examine urban simulation, single-lane dynamical systems, and human driver models. Address the critical challenge of scaling deep reinforcement learning in complex environments with combinatorial possibilities. Discover transfer learning across networks and zero-shot transfer techniques. Gain insights into counterfactual reasoning for societal systems and its potential applications in addressing the increasing pace of change and complexity in modern society.
Mixed Autonomy Traffic: A Reinforcement Learning Perspective
Overview
Syllabus
Intro
Counterfactual reasoning with • Motivation: Quantify impact of technology on societal systems • Pace of change & complexity is increasing
Years 2020 to 2049: Mixed autonomy Transportation in the US
Urban simulation
Axes of difficulty in mixed autonomy
Single-lane: dynamical system equil Human driver model
Challenge: combinatorial number of environn A critical challenge to scaling deep reinforcement learning
Transfer learning across networ
Zero-shot transfer
The road ahead: counterfactual reasoning for societa Motivation Quantity impact of technology on societal systems
Mixed Autonomy Traffic: A Reinforcement Learning Perspective
Taught by
Simons Institute
