Startup Proposal: Intelligent Traffic Signals
Traffic signals have always been run by timers set up by
traffic engineering professionals. They can be independent units,
possibly with pattern branches triggered by pedestrian requests or
lane sensors, or a number of units all under central computer
control. And we've all been stuck in frustrating traffic situations
where the signals just happened to go the worst possible ways at the
worst possible times.
What if traffic signals were networked intelligent devices with
iPhone-like microprocessor systems, sensors, and communications
links? What if they were able to continuously adjust and optimize
their timings for the most efficient possible traffic flow all on
their own? What if they could replace the monitoring and
controlling functions of municipal traffic control centers?
More efficient travel with less traffic congestion could likely save
billions of gallons of gasoline and tens of millions of tons of
exhaust emissions each year in the US alone. Less road
construction would be needed as existing roads are used more
The possibilities are enormous:
"Nest for Traffic Signals": Any intersection could
operate more efficiently by recognizing traffic patterns and
dynamically adjusting the timing to optimize for traffic
throughput (or any metric) as conditions change.
Working Together: Nearby smart signals will coordinate with
each other to optimize traffic flow with no central computer
involved. Connections to nearby units share data, communicate
needs, and dynamically adjust the phase offsets between the
signals for optimium traffic flow. When there is a congested
area, increase the green time to empty that area out while the
signals feeding the area reduce their green time. The
computational task increases with the number of units, but more
processors are crunching on it.
Map Data: Include the knowledge of local street topology
and capacity in these decisions. If there is a congested area,
adjust the timing to encourage alternate routes around it.
Modeling and Prediction: The traffic signals could also use
the networked data to anticipate upcoming traffic situations.
Potential congested areas could be emptied out before they become
Run Experiments: Operate with timings slightly tweaked,
collecting statistical data to find optimal traffic patterns for
each particular intersection. Share the experimental data with
other areas, compile statistics, and become smarter over time.
Synchronization: The synchronized traffic lights feature
(they call it "progression") currently requires central control of
all the units involved and an assumption of how quickly the cars
can get through. And this fails when those assumptions are not
met. Intelligent traffic signals could work together
automatically providing synchronization everywhere without central
control, dynamically adjusting the signal phases to accommodate a
wide variation in conditions. Optimum synchronization could be
maintained even with disruptions such as random pedestrian
Gridlock Conditions (cars stuck within intersections) could
be flagged and efficiently emptied out with a separate algorithm.
"This is a Job For a Computer": Municipal traffic centers
would no longer need to deal with the low level issues such as the
timings and offsets of individual signals, but instead describe
the overall goals, metrics, algorithms, street topology and note
special cases (holidays, parades, events, construction,
emergencies, etc.). Powerful intelligent traffic control would be
available to all towns and cities without a large investment.
Choose Your Metric: The performance metric can be
arbitrarily customized. "Optimize for exhaust emissions here,
but optimize for traffic throughput over here, and minimize
traffic down there."
It's a Platform: New parameters, algorithms, simulations,
and software could be uploaded at any time. The system would have
an open API for traffic management algorithms, so when some
researchers develop an improved or specialized algorithm, it can
be uploaded and put to use, with measurable results.
Simulation:As an alternative to algorithms, intelligent
traffic signals could perform continuous simulations fed by data
from nearby units and, like a classic computer chess program, look
several moves ahead and choose the best option.
Verifiable: The system could demonstrate results at any
time by reverting to the original signal timings and measuring the
Safe: The alorithms and data would be run on a separate
processor independant of the signal timing processor, with a
limited protocol between them. So any software problems would
leave the lights running with the most recent pattern. (There is
also the standard safety net called the Conflict Monitor
Unit that watches for inappropriate signal conditions and
brings the system down to blinky mode if anything is amiss.)
Extras: It could provide hooks to assist driverless cars. A
phone app could provide a pedestrian walk signal at any
intersection. Another phone app could charge the user for faster
access through traffic and provide revenue to the city.
Plug-In: The product will be delivered as replacement
plug-in controllers for current traffic signals.
A municipality could purchase smart traffic signals over time on a
replacement basis, similar to the switch to LED traffic lights.
Each new traffic signal will improve the performance of nearby smart
traffic signals so there is an incentive to add more.
There are at least 300,000 signaled intersections in the US.
International sales would be substantial. Cities and counties are
accustomed to spending $50,000 to $500,000 to purchase and install a
traffic signal. All sorts of services would be available by
subscription (statistical analysis, traffic control tools, algorithm
libraries, etc.). Annual revenues would be billions of dollars in
direct sales alone. This could be an enormous business that has a
profoundly positive effect on the environment.
I'm not seeing evidence of anybody currently doing anything like
this. There is a new development in traffic signals called
"Adaptive Traffic Control"
Adaptive Signal Control Technology), but all implementations so
far seem to be major projects involving a central computer system.
There are some experiments in the references below, certainly, but I
don't see any patents, or solid progress, or any in use, or anybody
approaching it the way I've described here.
Note that the city of Los Angeles recently finished a $400 million
project to connect all 4500 of their controlled intersections, with
video and sensor data, to a central computer facility staffed with
Times: To Fight Gridlock, Los Angeles Synchronizes Every Red
This would be a major project and would probably require a staff of
over 100 including experts in multiple areas (electrical
engineering, software engineering, simulation, artificial
intelligence, civil engineering, traffic engineers, etc.) and
venture funding of around $100 million. The market, world wide,
seems to be somewhere around $50 billion.
Zubillaga, et al:
Measuring the Complexity of Self-Organizing Traffic Lights
How AI Turns Traffic Lights Into Intelligent Agents
Design of Intelligent Traffic Light Controller Using Embedded
Carnegie Mellon University:
Smart Traffic Signals
Carnegie Mellon University:
Smart Signals — Pilot Study on Traffic Lights Reduces
Pollution, Traffic Clogs
US Department of Transportation:
Manual on Uniform Traffic Control Devices
Smarter programming of stoplights could improve efficiency of
How smart traffic signals may ease your commute
Choreographing the Dance of Traffic Lights
Ghena, et al:
Green Lights Forever: Analyzing the Security of Traffic Infrastructure
Boston Transportation Department:
The Benefits of Retiming/Rephasing Traffic Signals in the Back Bay