In a Bay Area field trial, a small fleet of vehicles used seconds‑ahead information to reduce hard braking and time spent stopped—building on lessons from CIRCLES and the 2022 “MegaVanderTest.”
UC Berkeley and Nissan researchers recently showed how coordinating drive patterns of a few vehicles can measurably smooth traffic for miles, making commutes safer and more efficient.
The demonstration, Cooperative Congestion Management (CCM) Phase II, was conducted on Interstate 680 in Contra Costa County. The team shared traffic flow information from a leading “probe” car, which helped following vehicles gently slow down before reaching a backup—cutting down on abrupt braking and stop‑and‑go waves over hundreds of real‑world miles.
“Human drivers create most phantom jams without meaning to—tiny brake taps cascade into stop‑and‑go waves,” said Alexandre Bayen, Professor of Electrical Engineering and Computer Sciences and Director of CITRIS and the Banatao Institute. “CCM borrows from our earlier CIRCLES work to show that even a handful of coordinated vehicles can stabilize flow in live traffic, using capabilities already on today’s cars.”
Why it matters
Stop‑and‑go traffic wastes time, fuel, and attention, and it increases rear‑end crash risk. CCM aims to shape traffic flow: a probe vehicle ahead shares near‑term traffic conditions so that vehicles 30–60 seconds behind can adjust speed early, creating a moving shock‑absorber that smooths the queue. The core idea—coordinating a small fraction of vehicles to benefit everyone—traces back to CIRCLES, including the 2022 open‑road field test in live traffic on Tennessee’s I‑24 (often referred to as the MegaVanderTest). There, researchers equipped 100 production vehicles with AI‑tuned cruise control to damp stop‑and‑go waves on I‑24.
What was tested on I‑680
Working with the Contra Costa Transportation Authority (CCTA), UC Berkeley and Nissan researchers integrated CCM software with production advanced driver‑assistance features and conducted supervised runs during the afternoon commute on I‑680. The team focused on key safety and comfort indicators: (1) hard‑braking events, (2) time spent stopped, (3) time‑to‑collision (TTC) margin, and (4) acceleration variability (interquartile range, IQR). Vehicles operated under a protocol that allowed drivers to take over at any time, with data logged for independent evaluation.
“With only a modest number of instrumented vehicles, seconds‑ahead targets fed into adaptive cruise control let the vehicles ease into queues earlier—reducing harsh braking and time spent stopped,” said Dr. Jonny Lee, a research scientist and project lead at UC Berkeley Institute of Transportation Studies. “Small automated adjustments by a few cars can stabilize flow for many.”
“A key challenge is balancing comfort and smoothing,” added Dr. Jerry Chou, senior researcher at Nissan’s Advanced Technology Center–Silicon Valley. “When the controller requested decelerations that felt too aggressive, professional operators were inclined to override. We tuned the targets and gains so CCM-enhanced adaptive cruise control could smooth flow while keeping speeds and following distance within socially-acceptable ranges.”
By the numbers (Phase II highlights)
-
Corridor: I‑680, San Francisco Bay Area (Contra Costa County)
-
Fleet size per run: one probe vehicle + 2–5 ACC‑controlled “control” cars
-
Penetration: comparable instantaneous penetration to larger trials, but not sustained for as long
-
Test exposure: ~600 miles of supervised driving
-
Observed outcomes when CCM control was active:
-
85% reduction in hard‑braking events
-
70% reduction in time spent stopped
-
Elimination of under 2-second time‑to‑collision
-
20% narrower acceleration interquartile range
-
Enabling tech: production‑grade cameras and radar from advanced driver‑assistance systems (ADAS); cellular connectivity for information sharing
Building on CIRCLES and the 2022 MegaVanderTest
CCM’s strategy aligns with insights from the CIRCLES Consortium’s open‑road experiments showing that coordinated vehicle behavior can dissipate phantom jams, improve fuel economy, and reduce emissions. The MegaVanderTest on I‑24 MOTION in Tennessee deployed about 100 AI‑equipped vehicles and demonstrated that small, well‑timed speed adjustments by a subset of cars can stabilize flow for the surrounding traffic.
How CCM Phase II differs: Unlike MegaVanderTest’s large, sustained fleet, CCM Phase II used a much smaller deployment—one probe vehicle plus 2–5 ACC‑controlled “control” cars per run. During each run, the instantaneous penetration in the local traffic stream was comparable, but not sustained for as long or as far. Even at this limited scale, the Bay Area trial produced substantial improvements, especially in safety‑oriented KPIs (hard braking, time stopped, TTC margin, acceleration IQR) rather than fuel‑consumption metrics.
Earlier roots include ring‑road experiments by CIRCLES co‑PIs and Berkeley’s reinforcement‑learning Flow simulator, which explored how small automated speed adjustments could dampen stop‑and‑go waves. CCM extends these ideas to a new corridor and hardware stack, emphasizing controller‑led ACC integration and deployability on existing vehicles.
What’s next
The team is exploring how to scale CCM using widely available connectivity and production ADAS, with emphasis on corridor‑level evaluation, safety validation, and mixed‑fleet operations in coordination with regional partners. Future work includes larger‑scale trials, continued safety analysis, and integration with roadside sensing and traffic management strategies.
“Cooperation is the point,” Bayen said. “Feeding downstream conditions into adaptive cruise control lets a small number of vehicles make gentle automated adjustments that ripple through traffic—reducing stop‑and‑go and improving safety.”
Project partners: UC Berkeley Institute of Transportation Studies; Nissan Advanced Technology Center–Silicon Valley; Davteq, Inc.; Contra Costa Transportation Authority (CCTA). Funding and support acknowledged from federal and regional programs.