Background
Stop-and-go waves, often called “phantom jams,” form on highways even without bottlenecks, lane closures, or crashes. They arise from the way human drivers react to the vehicles in front of them, and they waste large amounts of fuel and time. Earlier closed-course experiments showed that a small number of automated vehicles, driving with the right control strategy, can dampen these waves. The CIRCLES consortium was formed to take this finding from the test track to real highway traffic.
Problem statement and objectives
CIRCLES aims to advance mixed-autonomy traffic: the insertion of automated vehicles into human-driven traffic to control and optimize the flow. The project leverages deep reinforcement learning and optimal control to reduce energy consumption and eliminate stop-and-go waves, with a target of at least 10% energy savings for all vehicles on the road. The research plan has six parts:
-
Develop high-fidelity simulation tools that reproduce real traffic instabilities
-
Discover new multi-agent reinforcement learning techniques
-
Transfer learned policies to connected and automated vehicles
-
Develop, calibrate, and validate vehicle energy models
-
Combine sensing, control, and communication for traffic flow experiments
-
Build highway sensing infrastructure that observes the position of every vehicle on the road
