General

Fun Play: How Self-Driving and Human-Powered Cars Can Share the Road – News

Fun Play: How Self-Driving and Human-Powered Cars Can Share the Road - News
October 19, 2022

Just like when Model Ts traveled alongside horses and buggies, autonomous vehicles (AVs) and human-powered vehicles (HVs) will one day share the road. How to best manage the rise of AVs is the subject of a new policy letter from Carnegie Mellon University, “Mixed Autonomy Transport Era: Resilience and Autonomous Fleet Management.”

The debate continues over when AVs will dominate the streets, but one of the letter’s authors, Carlee Joe-Wongsaid, “Once AVs start implementing, there’s probably no going back. So there’s a need to talk about policies now, study them in depth, and get them right by the time AVs arrive.”

Joe-Wong, the Robert E. Doherty Professor of Career Development of Electrical and computer engineeringthe research team behind the brief said, “What’s different when you have AVs in the mix compared to when you only have HVs? We realized that one of the main differences between AVs and HVs is that AVs are altruistic and are HVs selfish.”

AVs can anticipate what is about to happen and redirect themselves, for example in the event of road works or an accident. Programmed to operate safely and follow rules, AVs can take altruistic actions that benefit other vehicles and not just themselves. People in a hurry may not be so generous with their time.

The price of selfish driving becomes apparent when examining the flow of traffic. When selfish cars drive in and out of a traffic system, the system will eventually reach an equilibrium, a state of equilibrium, but the traffic cannot flow as efficiently as it could. This way, balance can be achieved when traffic rages bumper to bumper. “Sometimes the balance is far from optimal,” Joe-Wong says.

The researchers believe that altruism can improve traffic flow by avoiding sub-optimal equilibria, and not everyone has to be nice to improve travel times. In simulations, altruistic states come into play when AVs make up 20% to 50% of the vehicles on the road. The report suggests ways to reward altruism, including incentives such as toll waivers and parking discounts.

Finding the best operational policies for AV fleets is another topic covered in the report. AVs can operate synchronously, but controlling thousands of AVs centrally will lead to computational problems and communication delays. The researchers aim to strike a balance between centralized and decentralized policies using reinforcement learning, a machine learning training method.

The engineers have considered how AVs make decisions. How does machine learning help with this process and what kinds of decisions have the greatest impact? According to Joe-Wong, “In some circumstances you really need reinforcement learning intelligence, but in other circumstances reinforcement learning just tells you to do what you probably would have done anyway.”

The team proposes that fleet managers train models to manage AV fleets locally. As new traffic patterns emerge, the models are updated, mainly to steer people away from incidents. However, if traffic continues to flow, fewer updates are needed, reducing communication between AVs on the road and AVs reporting to a centralized server.

The final problem the researchers examined was how to deal with traffic congestion and avoid cascading failures.

Working in optimal balance, applying reinforcement learning, and having a greater proportion of cooperating AVs will reduce congestion. However, to address the failure of cascading failures, the researchers took into account other modes of transport found in urban networks. They added bus, metro, rail and bicycle sharing systems to their models, and were able to show that if passengers were adapted between different modes of transport, it would maximize the use of the entire network and prevent it from becoming overloaded and malfunctioning.

Based on their findings, the team recommended that when planning agencies establish traffic flow redistribution policies for AVs, they consider how to integrate multiple interdependent transportation systems to keep people moving.

In the age of mixed autonomy, altruistic AVs can act as coordinators that keep traffic flowing by provoking positive actions from HVs. While it will take some time for AVs to outsize human-powered vehicles, all drivers will notice improved traffic flows with only partial adaptation of AVs.

In addition to Joe-Wong, investigators of this letter include: Osman Yağan, research professor, and I-Cheng Lin, Ph.D. candidate, both in the Department of Electrical and Computer Engineering. This work was supported by the National University Transportation Center for Improving Mobility (Mobility21) at Carnegie Mellon University.