CSAIL Publications and Digital Archive header
bullet Technical Reports bullet Work Products bullet Research Abstracts bullet Historical Collections bullet

link to publications.csail.mit.edu link to www.csail.mit.edu horizontal line

 

Research Abstracts - 2006
horizontal line

horizontal line

vertical line
vertical line

CarTel: A Mobile Sensor Computing System

Hari Balakrishnan, Samuel Madden, Vladimir Bychkovsky, Kevin Chen, Waseem Daher, Michel Goraczko, Hongyi Hu, Bret Hull, Allen Miu & Eugene Shih

Project Overview

Over the past few years, impressive advances in wireless networking and embedded computing have led to the "first generation" of wireless sensor networks [1,2]. In general, these are for monitoring or tracking applications characterized by low data rates and static deployments. We believe that the "second generation" will include mobile sensor networks and high data-rate sensor networks. With hundreds of millions of vehicles and over a billion cell phone-equipped people in the world, cars and humans may turn out to be the carriers of the world's largest and most dynamic sensor networks in the coming years.

How does one combine wireless network protocols and data management software in a world where sensors are mobile (e.g., on cars), collect media-rich sensor data from a variety of sensors (e.g., cameras, vibration sensors, On-Board Diagnostic (OBD-II) data from the car's internal sensor network, Wi-Fi and cellular network monitors, etc.) and may only be intermittently connected to servers on the Internet? The CarTel project is investigating these questions by designing, building, and deploying real-world mobile sensor network systems. Our primary (and current) focus is on automotive applications, which include:

  1. Traffic monitoring and route planning. Suppose we tracked the location of every car (suitably anonymized) once per second using GPS. We can use this information to develop statistical models of traffic delays at various times of day on different road segments. Suppose you want to leave your home for the airport to catch a flight at 8:00 am. Which of the four different routes to the airport should you take? The statistics of delays between 6:00 am and 7:30 am can be used to develop a smarter route planning system than what is available today. Moreover, models of past delays can be incorporated with real-time information about accidents and weather to provide routes that can save time.
  2. Preventive maintenance and diagnostics of cars. By tapping into the on-board sensors using the standard CAN (controller area network) interface, and by attaching a variety of external sensors, we can monitor and report internal performance characteristics such as emissions, gas mileage, tire pressure, suspension health, etc. These reports can be combined with historical data, highlighting long-term changes in a car's internals, and correlating a given car's information with other cars of the same vintage to detect anomalies in a car's performance.
  3. Civil and environmental monitoring. When equipped with additional sensors to sense vibration and other conditions, cars can act as excellent ``probes'' to sense environmental and road conditions. Assessing and reporting road surface conditions such as pollution levels, air quality, potholes, oil spills, flooding, etc. can potentially be done effectively using CarTel.
  4. Wireless network monitoring. Such a system could provide a scalable, natural service to monitor the quality of various wireless networks that are around us. Examples include WiFi deployments and various cellular wireless infrastructures.
  5. Cars as ``data mules''. An increasing number of sensor networks are being deployed at remote locations for applications ranging from civil engineering (e.g., pipeline monitoring in cities) to environmental monitoring (e.g., monitoring pollution levels) to weather monitoring. A significant challenge in these deployments concerns the delivery of the data gathered by these networks to a central station. A network of cars can act as ``data mules'', carrying the data collected by sensor networks to their eventual destination.
  6. Cars as peer-to-peer (P2P) networks. With enough density, cars on the road can communicate with another, enabling a variety of geographic P2P applications (e.g., games or streaming audio) as well as sharing of information about available parking spaces, nearby traffic cops, or potential carpool routes.
CarTel Objectives

The CarTel project has three research thrusts:

  1. CafNet, an opportunistic, delay-tolerant carry-and-forward network that integrates opportunistic network discovery, message routing, and storage. Data may be carried by "mules" between origin and destination, which store data and move them physically toward their destination.
  2. CarTelDB, a delay-tolerant continuous query processor, which is the data management and declarative query processing framework for the system.
  3. The CarTel AutoPortal, a spatial map-based user interface that users interact with, which provides a visual way to interact with the system.

CafNet and CarTelDB are described in more detail in accompanying research abstracts.

Initial Progress

Our initial efforts have focused on deploying a small number of remote nodes in vehicles and building a portal prototype. Pictured below is the hardware that we install in each car. This includes placing a small, embedded Linux computer equipped with a variety of sensors, including GPS, and a Wi-Fi network interface in the trunks of our deployment vehicles. We also have a connection to the car's standard on-board diagnostic (OBD) interface to monitor engine performance. As these cars drive around they collect sensor data and buffer it using CarTelDB until CafNet finds a data delivery "conduit" (such as Wi-Fi) to forward data over.

cartel node hw cartel node hw internals vehicle placement

All sensor data is collected and archived at a central server and exposed to the user through the CarTel AutoPortal. The screen shots below show a prototype interface that allows users to view routes they have taken. Each route has "data overlays" that correspond to the sensor data being collect by the remote nodes. For example, users can overlay engine performance metrics on these routes to visualize changes as a function of time and geography.

Screen shot of the AutoPortal prototype Screen shot of the AutoPortal prototype

For more details, see the CarTel site.

Related Work

The CarTel mobile sensor computing system builds on ideas from many research areas.

There has been much written about deploying sensor systems for environmental monitoring and data collection [1,2]. For the most part, these systems are focused on low-data rate sensing and power management issues.

In the area of query processing, our system extends many of the ideas pioneered in such continuous query engines as [3,4,5]. However, unlike most traditional systems, CarTelDB pushes into new areas by treating intermittent connectivity as a fundamental property of the system that must be exploited rather than masked or considered a failure.

In the area of networking, there have been several proposals for delay tolerant network stacks including [6]. In addition, there has been some initial work that explores sending data in a mobile context using opportunistic Wi-Fi [7] as well as through occasionally connected data mules [8]. CafNet applies and extends these ideas using a novel application programming interface and works over a variety of networking technologies.

References:

[1] G. Tolle, et al. A macroscope in the redwoods. In SenSys, 2005.

[2] N. Xu, et al. A wireless sensor network for structural monitoring. In SenSys, 2004.

[3] D. J. Abadi, Y. Ahmad, M. Balazinska, U. Cetintemel, M. Cherniack, J.-H. Hwang, W. Lindner, A. Maskey, N. Tatbul, Y. Xing, and S. Zdonik. Design issues for second generation stream processing engines. In Proc. of the Conference for Innovative Database Research (CIDR), Asilomar, CA, Jan. 2005.

[4] M. Balazinska, H. Balakrishnan, S. Madden, and M. Stonebraker. Fault-tolerance in the borealis distributed stream processing system. In Proc. of the 2005 ACM SIGMOD International Conference on Management of Data, pages 13-24, Baltimore, MD, 2005.

[5] S. Chandrasekaran, O. Cooper, A. Deshpande, M. J. Franklin, J. M. Hellerstein, W. Hong, S. Krishnamurthy, S. R. Madden, V. Raman, F. Reiss, and M. A. Shah. TelegraphCQ: Continuous dataflow processing for an uncertain world. In Proceedings of First Annual Conference on Innovative Database Research (CIDR), 2003.

[6] K. Fall. A Delay Tolerant Networking Architecture for Challenged Internets. Proc. SIGCOMM 2003, Aug. 2003

[7] J. Ott and D. Kutscher. Drive-thru Internet: IEEE 802.11b for Automobile Users. In Proc. IEEE INFO- COM, Hong Kong, March 2004.

[8] P.Juang, H.Oki, Y.Wang, M.Martonosi, L.S.Peh, and D.Rubenstein. Energy-Efficient Computing for Wildlife Tracking: Design Tradeoffs and Early Experiences with ZebraNet. Proceedings of ASPLOS-X, San Jose, October 2002.

Acknowledgments

This work is funded in part by Quanta Corporation and by the National Science Foundation under Award Number CNS-0205445.

vertical line
vertical line
 
horizontal line

MIT logo Computer Science and Artificial Intelligence Laboratory (CSAIL)
The Stata Center, Building 32 - 32 Vassar Street - Cambridge, MA 02139 - USA
tel:+1-617-253-0073 - publications@csail.mit.edu