Tu1A15 - Bridging the Gap in Autonomous Vehicle Controls in Mixed Traffic

Description

The objective of this tutorial is to identify gaps and challenges arising from connected and automated vehicles (CAV) controls when CAVs have to drive alongside humans for the next few decades or so. Although autonomous driving technologies are promised to improve traffic safety and efficiency, the presence of human actors such as drivers, pedestrians, cyclists could jeopardize and even compromise the anticipated performances of CAVs. It is urgent to develop knowledge, models, and control methods to accommodate such uncertainty and instability brought forth by various human actors on public roads. Through this tutorial workshop, we aim to bring experts from both academia and industry who will brainstorm potential solutions and lay out a feasible and collaborative path to the ultimate success of full connectivity and autonomy.

One focal point of the discussion is how existing control models, methods, and algorithms should be modified to accommodate challenges induced by humans. Discussions will be contingent upon vehicle automation levels, from Level 1 (Driver assistance) to Level 5 (Full automation), as well as communication technologies, including Dedicated Short Range Communication (DSRC, recommended by US DOT) and 5G/wireless communication. Industry and academia may prioritize different technologies while developing understanding and controls for CAVs. This tutorial aims to achieve a convergent discussion.

Presenters

Title: Automated Driving in Mixed Traffic
Authors: Huei Peng (Professor at University of Michigan) & Shaobing Xu (Assistant Research Scientist at University of Michigan)
Time: 10:15-11:15 AM EST

In this talk, we will cover four topics: path planning, motion control, prediction of other driver's behavior, and the concept of driving with roadmanship. Key motion control challenges include high time delay and tight road curvature. A delay-and-dynamics-aware preview control algorithm will be presented, which generates closed-form steering compensation for the delay and feedforward steering for future path curvatures in an integrated fashion. For motion planning, we will present a scalable algorithm stack that adapts to diverse traffic scenarios, which resolves the planning problem with bounded computing time and near-optimality. Finally, two concepts related to predicting other vehicles' future motions and plan for the AV not only to drive safely, but also with good roadmanship will be explained. Deployments and experiments using the Mcity self-driving car fleet will be presented.

Title: Traffic Flow Smoothing at Scale
Authors: Daniel Work (Associate Professor at Vanderbilt University)
Time: 11:15-11:35 AM EST

The majority of the best-selling cars in the US are now available with SAE level-one automated driving features such as adaptive cruise control. As the penetration rate of these vehicles grows on the roadways, it is now possible to consider controlling the bulk human-piloted traffic flow by carefully designing these driver-assist features. This talk will discuss modeling, simulation, and field demonstration advancements that are needed to control automated vehicles to stabilize traffic flow at scale. Prior work on a closed course established that automated vehicles can eliminate human-generated phantom traffic jams that seemingly occur without cause, reducing fuel consumption by up to 40%. The talk will highlight the research challenges and progress towards demonstrating traffic flow smoothing with a fleet of connected and automated vehicles on the I-24 Smart Corridor in Tennessee, as part of the CIRCLES Consortium.

Title: Learning and Influencing Routing Preferences on Mixed-Autonomy Traffic Networks
Authors: Dorsa Sadigh (Assistant Professor at Stanford)
Time: 11:35-11:55 AM EST

Panel Discussion
Time: 12:00-12:15 PM EST

Schedule

The planned schedule is as follows: