 MIT researchers fight gridlock with Linux |
|
At the Massachusetts Institute of Technology (MIT), researchers are testing a Linux-based automotive telematics system intended to reduce traffic congestion. CarTel is a distributed, GPS-enabled mobile sensor network that uses WiFi "opportunistically" to exploit brief windows of coverage to update a central traffic analysis program.
|
| |
|
|
The CarTel system is used to
test a growing number of automotive-related research projects at MIT's
Computer Science and Artificial Intelligence Laboratory (CSAIL). In
some ways, the system is similar to the recently announced  Dash Express
navigation system, in that it uses embedded Linux, GPS, and WiFi, and
links up to a central route analysis system. Unlike Dash, however,
CarTel currently lacks a display or real-time navigation and mapping
features. Dash, meanwhile, offers additional location-based services
and incorporates a cellular modem (850MHz triband GSM) to provide a
continuous connection, switching to WiFi when available.
Some experiments using the CarTel
design have used a cellular modem, says Associate Professor Sam Madden
in an interview. Yet, the main goal is to find out how much can be
achieved with an intermittent communications system. Cellular service
would add significantly to the operating cost of the system.
Madden, who leads the project
along with Professor Hari Balakrishnan, says CarTel's goal is to
establish a flexible, affordable platform for diverse
automotive-related research projects. These include testing
continuous-stream querying, exploring issues of mobile, intermittent
WiFi, and researching fleet management applications, driver safety
technology, and identifying pothole location for maintenance planning.
The
main focus, however, is to develop sophisticated route-selection
algorithms to address the problem of congested traffic. "Everyone
agrees that traffic is a pain," says Madden, "so avoiding it is good.
Hopefully this will have some direct benefit to users." Improving
routing could have a significant impact on commute time, says Madden,
which in turn would reduce fuel consumption and pollution.
As
a test deployment, the CarTel team installed Linux-based computers and
sensor kits in 30 limousines from a Cambridge, Mass.-based livery
service called PlanetTran. For the last year, the hybrid-based limos
have been driving around collecting data about Boston-area traffic
conditions, as well as information about potholes, WiFi availability,
and the health of the cars themselves.
The CarTel computer is equipped
with an OBD-II (on-board diagnostics II) interface to the car's sensor
network. This lets it record speed, mileage, emissions, RPMs, and other
information. In addition, the car computer hooks up to additional
sensors, depending on the experiment, such as cameras or three-way
accelerometers to measure and map potholes. Other possibilities include
noise and pollution detectors.
Inside CarTel
The current hardware platform for CarTel is a box about 7 x 5 x 1 inches equipped with a Soekris net4801
single-board computer (SBC) running Linux 2.6. The computer has a
"586-class" CPU running at 266MHz, with 128MB of RAM, and 1GB (or more)
of flash. The CarTel box is equipped with two serial ports for
collecting sensor data, plus a USB 1.1 port, and a WiFi card plugged
into a miniPCI slot. A commodity GPS unit is connected via one USB
port. A Bluetooth dongle connects to the other USB port, allowing
connections from a mobile phone. For telematics data, the system uses
an OBD-to-serial adapter from ScanTool.net.
The MIT team has developed some
innovative technology to get the most out of fleeting, off-and-on
encounters with WiFi access points. First, it has developed a WiFi
communications protocol called EasyWiFi that is optimized for brief
encounters. EasyWiFi takes only about a hundred milliseconds to
establish a mobile wireless connection, says Madden, instead of a more
typical five to ten seconds. The modifications are only on the client
side, so the protocols can work with any WiFi connection.
Other
communication protocols handle variable and intermittent connectivity.
A "dpipe" (delay-tolerant pipe) transport programming abstraction
developed by postdoc Jakob Eriksson allows producer and consumer nodes
to reliably transport data across an intermittently connected network.
The client-server dpipe connection is optimized for start-and-stop
delivery and for situations when the IP address of an end-point changes
frequently.
The dpipe uses a modified delay that can maintain
its status even when the session is dropped, says Madden. "When
connectivity is availability again, it pops the data out the WiFi
connection to the server where it stores it in memory," he adds.
A
longer term store-and-forward technology, called CafNet, (carry and
forward network) employs protocols that enable cars to serve as data
mules, delivering data between nodes that are not connected via
telecommunications. This technology could be used, for example, in
military, mining, agricultural or scientific applications that are
spread out over vast distances, delivering data from sensor networks
deployed in the field to Internet servers. Each sensor would require
only short-range WiFi connectivity, thereby lowering costs and easing
maintenance.
Real-time queries and what-if route algorithms
Distributed
sensor data is processed by a portal application that is built around a
Linux- and SQL-based stream-processing query application called ICEDB.
This delay-tolerant, continuous query processor enables applications
running on the portal to issue queries via an API, or transfer a
sequence of data between nodes using dpipe. Queries can specify what
sensor data is needed and at what rate, and how the data should be
sub-sampled, filtered, summarized, and prioritized.
Applications
can also query the portal's relational database for analysis. These
"snapshot" queries work from available data and don't need to wait
synchronously for complete results. All this underlying complexity is
shielded from the developer, says Madden, so it appears like a standard
SQL relational database.
The ICEDB database was modified from
a similar "TinyDB" database that Madden developed years ago for the
TinyOS operating system. TinyDB was used to query continuous-feed data
from sensor networks that used sensor motes from Crossbow Technologies.
"We
made a conscious decision to move to Linux because TinyOS was not as
easy to work with," says Madden. "With Linux, there are also a huge
number of people developing device drivers, and our graduate students
already know how to develop with it."
The CarTel portal provides a
geo-spatial data visualization system based on Google Maps that stores
and marks sensor data using GPS coordinates. The portal organizes data
as "traces," linear sets of sensor readings collected during a drive.
Users can query the system using a graphical query interface to select
traces, and then visualize the traces combined with various summaries,
statistics, and maps. The portal allows queries both on the driver's
own history, as well as the aggregate driving history of other drivers.
Users
could include fleet directors, city planning officials, or the drivers
themselves, logging in either from a WiFi-enabled laptop in the car or
later at home. For example, drivers can see congested areas of their
commutes marked in red, orange, and yellow segments, with greens
indicating average speeds, and the rare blues indicating clear sailing.
The current focus of the project is in developing algorithms
that run on top of the portal application to help drivers plot the best
route at a given time. For example, the team's MyRoute project includes
applications that model delays observed on road segments as statistical
distributions. Various algorithms then use these to compute optimal
routes for different times of the day.
"Instead of asking the
shortest time or shortest distance from point A to point B, you ask
what route should be taken, say, for the highest probability of getting
to the airport by a certain time depending on the time selected," says
Madden.
Meanwhile, additional research is underway using CarTel
to improve fleet management and fine-tune street repair schedules based
on identifying the deepest potholes that cars are hitting most
frequently. Other research will investigate using telematics data to
warn drivers of potential safety hazards or maintenance needs.
With
the Soekris net4801 expected to reach end-of-life later this year, the
CarTel group is planning to move to another Linux-based design, hoping
to reduce the current over-$400 pricetag for the complete system. The
plan is to expand the network to 1,000 vehicles, which will help
improve the granularity of collected data for better route planning.
Other potential lines of inquiry include peer-to-peer WiFi
communications between cars to sense upcoming traffic congestion or to
assist in onboard safety systems.
Last May, a U.S. government- and industry-led coalition called the Vehicle Infrastructure Integration Consortium (VII-C)
was established, aiming to equip every car and roadside in America over
the next decade with wirelessly connected Linux-based computers. The
goal is to lower driver death rates, reduce traffic jams, and
media-enable cars before 2017. The VII-C is funded by the U.S.
Department of Transportation (DOT), and includes seven vehicle
manufacturers already involved in the U.S. DOT's Intelligent Vehicle
Initiative.
More information on the MIT CarTel project can be found here.
Story by: Eric Brown Original story can be found at www.linuxdevices.com/news/NS8559536444.html
|
|
Updated Tuesday, February 05, 2008 Written by Admin 906 reads |
|
The Un-X Crew
Printerfriendly display 
