Roughly 40,000 Americans die every year in automobile accidents. Thousands more are maimed and injured. Many of these accidents are not caused by alcohol, speed, or reckless driving, but by simple driver error. For example, even if drunk driving were totally eliminated, cutting car fatalities roughly in half, automobile accidents would still be the leading cause of accidental death in this country.
Let me illustrate this point by personal example. My two most serious car accidents both occured while I was stone cold sober. One accident happened because I was tired and fell asleep at the wheel. Fortunately, it was just after dawn, there were few cars on the highway, and though my car spun out of control at around 60 mph, it didn’t flip over and came to a stop in a drainage ditch without hitting anything or anyone. In the second accident, I was making a left turn at a traffic light with a green arrow. An elderly gentleman ran the opposing stop light and broadsided me. Though I was not at fault, I can honestly say that if I’d bothered to look around, I’d have seen the other car and been able to stop in time. Instead, I was coming home from work, doing the same thing I did every other day at 5 PM, probably more concerned about finding a decent song on the radio, so when the light turned green, I just hit the gas and went.
Automobiles are simply dangerous. Air bags, lower speed limits, and mandatory seatbelt laws have reduced the fatalities but do not offer any ultimate solutions. Undoubtedly, drivers need to constantly remind themselves how dangerous this everyday activity is, and reduce distractions such as fatigue, intoxication, and car phones. Operating an automobile requires long periods of boring, repetitive work, interrupted rarely by unannounced moments requiring instant, intensive concentration. Humans by nature are not well suited to this kind of task, but it’s just the kind of thing computers excel at, so to significantly reduce automobile fatalities, computerizing the operation of cars has to be seriously considered.
Proposals for computerized automobiles have been around for some time. TRW prepared an story imagining the possibility of a computerized automobile system by 2012. In 1997, a modified stretch of California’s Interstate 15 served as the testbed for a series of demonstrations with automated cars and buses. The U.S. Department of Transportation has an Intelligent Transportation Systems Joint Program Office that coordinates many of these efforts. The Intelligent Transportation Society of America (ITS America) organizes conferences, maintains a website, and publishes a regular newsletter. Unfortunately, the goal is still distant, and many current ITS efforts are focused on programs such as more sophisticated traffic signals and alleviating congestion with automatic toll collection. The DOT-funded National Automated Highway Systems Consortium, which oversaw the I-15 tests and intended to develop a prototype system by 2002, has been terminated.
The highway in California was modified by placing magnetics in the roadway, which the vehicles then followed. Much important work has also been done on vehicles operating on unmodified roadways. CMU’s Robotics Institute developed a series of vehicles (the Navlabs) which drove from Pittsburg, PA to San Diego, CA under computer control for 98% of the trip. Navlab would make an excellent starting point for a smart car of the future, since CMU has already developed a controller system to operate the car, a standard API for operating the controller, and a simulation environment for testing new controller programs.
Some look to large corporations like GM or Toyota to design the smart car of the future. Others expect the initiative to come from the U.S. federal government. According to the U.S. Department of Transportation’s Intelligent Vehicle Initiative Governance Structure, the Enabling R&D group is only open to “vehicle OEMs with a World Manufacturer Identifier… and will require contribution of substantial financial resources” As a free software aficionado, I’d rather see a initiative to build an open, co-operative system in which governments, large companies, small organizations, and individuals can all contribute. The ability of the Internet community to develop complex, open source software systems, such as Linux, demonstrates the feasibility of using the Internet as a basis for collaboration.
To minimize the infrastructure requirements, the system would have to interoperate with ordinary cars on unmodified highways, basically Navlab’s approach to the problem. Much of the hardware required to support an automated car is already available:
Computer platform. The modern laptop computer seems well-suited to support a future smart car. It offers ample processing power and disk space, can operate off 12 VDC power, and is well standardized. PCMCIA cards provide a convenient and standard hardware interface. Slight modifications, such as a detached display to be placed on the dashboard, wouldn’t be difficult to implement.
Radiolocation. GPS (Global Positioning System) can’t provide enough accuracy to locate a car within a lane, but can locate a car within a dozen meters or so. Furthermore, GPS technology is mature, readily available at low cost, and easily integrated with existing computer technology. For example, Premier Electronics markets the SatNav GPS Receiver, a PCMCIA GPS receiver.
Communications. For long range communications, cellular telephones and cellular modems are expensive, but well understood and widely deployed. Upon locating itself with GPS, the car’s computer could dial into a server and download a database for the surrounding area. For short range communications, between nearby vehicles and traffic signals, the IEEE 802.11 wireless LAN standard is newer, but available as off-the-shelf, unlicensed products that work well within roughly a hundred meters.
Video capture. Small video cameras are commonly connected to video capture cards, providing a ready base for visual sensor systems. Navlab’s No Hands Across America demonstration relied heavily on their video-based RALPH system to follow highway markings. PCMCIA video capture cards can be connected to off-the-shelf miniature CCD cameras to provide a readily available video capability.
Radar. No really adequate automobile-based radar system exists today. Advances in microstrip fabrication technology, such as the ready, cheap availability of dielectric resonant oscillators, allows gigahertz-wavelength devices to be fabricated on a conventional PC board. It should now be possible to mass produce a low cost (sub $500) radar system to mount on an automobile and scan for other cars and pedestrians within a hundred meters. If you don’t believe this, check out my essay Guardian Alert: How it works for a description of a simple radar system and an outline of how it could be adapted for vehicular use.
Vehicle control. The development of Linux would have been impossible without the IBM PC – a standard, widely available hardware platform that software designers across the planet had ready access to. Likewise, to build an open source smart car, a standard software API needs to be developed for issuing commands like “drive forward at 20 mph”, “right turn 10 degrees”, “stop”. At least one model hardware implementation needs to be made readily available in kit form, probably using a USB serial interface, and simple enough that an average auto mechanic could install it on an automatic transmission car.
The most important thing now is to collect together the available technology, publish it on a website, and launch a collaborative effort to synergize the talents of the Internet community. An open-source version of CMU’s Navlab would make an excellent start.