The next generation of cars will be able to drive themselves without human intervention

frozenviolet

The next generation of cars will be able to drive themselves without human intervention [Copy link]

The team behind Stanley is still working on it; the car just won the 2005 Defense Research Projects Agency self-driving car competition across 132 miles of the Nevada desert. By 2008, the Stanford team will be taking their self-driving car on an interstate trip.

"The next milestone we are heading to now is to demonstrate the possibility of autonomous driving in traffic," said Professor Sebastian Thrun, leader of the Stanford Artificial Intelligence Laboratory and the university competition team. "Our goal at Stanford is to be able to drive the car from San Francisco to Los Angeles without human intervention in the next few years." The

entire drive will take several hours, Thrun said, "and will go through a variety of traffic conditions, from city driving to crowded highways to long distances across the vast interstate. We think that this is enough to demonstrate the ability of autonomous driving."

According to Thrun, the 2005 challenge "has answered the question of whether you can drive autonomously in an open desert at moderate speeds, which is essentially nothing like real driving. So the next question is, can you drive autonomously at higher speeds in real traffic?"

Semiconductor design engineers, automotive engineers, software and middleware experts all say "yes", and autonomous cars (cars that drive themselves without human intervention) will soon be on the market. "The day when consumers can have autonomous cars is not far away."

BMW's "active drive" technology can adjust the gear ratio according to the speed of the car's self-driving.

Sensors for such vehicles are already available, and semiconductor manufacturers are following quickly. "We believe that autonomous vehicles will be the main growth area for automotive electronics in the next few years," said Peter Schulmeyer, director of strategy and marketing for transportation products at Freescale Semiconductor. "The ultimate goal of autonomous vehicles is -- from today's point of view -- to gradually achieve the automation of car driving, meaning that the car itself can drive itself without the driver or passengers intervening in any equipment."

Stanford's Thrun predicts that it will take at least 30 years for cars to achieve full autonomous driving -- not just lane departure prevention on the highway. But there are many milestones between the future and reality, such as autonomous military escorts and a large number of convenience and safety features that will be gradually promoted to commercial and consumer vehicles, providing them with varying degrees of automotive autonomy.

"The development of automotive electronics has stimulated a great development in car design," said Mike Williams, chief analyst of Gartner Dataquest's global semiconductor group. "The global automotive semiconductor market exceeded $16.7 billion in 2005. The current innovation of automotive electronics driven by safety will enable future cars to achieve autonomous driving capabilities."

Freescale's Schulmeyer believes that collision avoidance will be a must-have goal for fully autonomous driving, and various innovations in the car will enhance automation. On this development path, he believes that it will start with simple comfort features such as automatic windshield wipers, and then safety systems such as anti-lock ABS, and then vehicle stability systems and adaptive cruise control. "These will be the necessary systems to achieve collision avoidance, rather than just relying on preload sensors and airbags to minimize damage."

Schulmeyer predicts that there will be three different stages of development for automatic safety systems. First, adaptive cruise control will use radar technology; second, more active safety systems, such as emergency braking in the event of an unavoidable collision, will minimize damage with maximum braking force; third, full collision avoidance systems can manipulate the car according to the obstacles encountered to prevent any collision.

Proven feasibility

Since the 1960s, researchers have been exploring various approaches to making cars autonomous, including fully autonomous cars like Stanley, where the vehicle's software makes the choices; and central control, where a master traffic computer signals equivalent slave vehicle computers on an automated highway. Many of these technologies have been proven feasible in field tests on highways, as electronics have shrunk from large footprints to microchips. Highway-specific technologies can now be easily adapted to the dashboards of modern commercial and consumer vehicles, so why can't cars drive themselves?

One reason is that the vehicle needs to have "drive-by-wire" capability, which means that all mechanical linkages, from the accelerator, transmission, brakes to the steering wheel, need to be controlled by computer-activated electrical servos. For example, computers directly control ABS systems, and drive-by-wire is now being installed in many cars. BMW's active steering technology uses computers to adjust the angle of the steering wheel independently of the angle of the car's front wheels to compensate for crosswinds.

BMW, Daimler-Chrysler, General Motors, Honda, Mercedes, Toyota and Volkswagen all plan to build drive-by-wire features into their cars for convenience and safety reasons, but not fully autonomous cars. "People don't just let the car navigate, steer and brake, take you to your destination and that's it," analyst Williams said, "because then the fun of driving is gone, and people don't necessarily need the car to drive itself."

So, "even if you admit that autonomous cars will be popular, for people who love driving, they still want to drive themselves," said Stanford's Thrun, who is an owner of sports cars himself, "because I love driving."

If you can switch to autonomous mode, he said, "you will be more productive, you can read the newspaper, sleep or answer e-mail, even watch a movie." And such a car will give peace of mind to elderly drivers who may have given up their driver's licenses.

According to Thrun, fully autonomous cars make automatic parking safer. In fact, some automakers have already demonstrated fully automatic parallel parking. "Even if the self-parking is in a static environment, the requirement for full computer control of the steering, brakes, transmission and accelerator is the same," Thrun said. "We have adaptive cruise control, lane departure systems, ABS -- everything is technically controlled by the computer in the car."

Other current safety features are aimed at enhancing driver awareness for future fully autonomous driving. For example, current overhead night vision systems can project long-distance road conditions detected by infrared sensors through fog onto the windshield. In the future, autonomous vehicles can use these sensors to identify objects matched to GPS maps.

“There’s a lot of attention paid to a driver alert system that gives you data from sensors that allow you to perceive the road in some way,” Thrun said. “For example, night vision systems send infrared light out of the car and always have high penetration because the eye can’t see infrared. But with a camera that collects infrared light, you can see the entire scene, which greatly helps people see the road conditions through the dark.”

He pointed out that radar emergency warning systems can advise you to slow down because they see obstacles through the fog, and new driver assistance systems can let you distinguish the types of obstacles. “This is technology that makes driving safer for everyone,” he emphasized.

Freescale's Schulmeyer believes that fully autonomous vehicles will evolve from fail-safe devices to automatic fault-tolerant systems, just as the aviation industry has developed. "Currently, fail-safe systems have an option to take immediate disengagement measures when an error is detected; for example, when the airbag or brake system turns itself off, a display on the dashboard lights up to indicate a self-test failure," Schulmeyer said. "Automatic fault-tolerant systems, on the other hand, have no safety bit."

If an error is detected during operation, for example, "you might not be able to turn the steering wheel if it is detected to be faulty," he said. "For safety and reliability, when moving from fail-safe devices to automatic fault-tolerant systems, you need fault-tolerant communication between subsystems, you need redundancy of critical components, and your software stability must meet very strict safety standards."

To encourage automakers to innovate fault-tolerant communication systems, Freescale Semiconductor has created a consortium called FlexRay based on its free real-time software. Currently, supporting manufacturers include: Audi, BMW, Bosch, DaimlerChrysler, Ford, Freescale, General Motors, Philips, Siemens and Volkswagen.

Similarly, Real-Time Innovations (RTI) provides a network data distribution system (NDDS) called middleware to handle real-time communication between the many processors in the car.

"NDDS middleware acts as an arbitrator between sensors and high-level software for recognition. It distributes data from sensors to algorithms that anticipate real-time constraints," said Rajive Joshi, chief engineer at RTI.

As middleware that connects data to processing nodes, NDDS adopts symmetrical back-to-back distribution, eliminating the need for a central server, and all nodes look the same to NDSS.

"NDDS feeds the algorithm the sensor data it needs, only when it needs it, no matter how many sensors there are or how many algorithms are accessing the same sensor," Joshi said. "The sensors are the 'publishers' of the data stream, and the algorithms are the subscribers; if the sensors are redundant, then when data is unavailable from one sensor, NDDS replaces the data from another sensor in real time without interruption. Even if no new 'published' sensor data is available to the real-time subscriber, NDDS notifies the subscriber that an unpredictable event has occurred, so its algorithm can replace the inserted data based on its missing data."

In addition to sensor fusion, NDDS also moves data navigation and stimulation systems. Joshi said that last year the U.S. Department of Defense standardized NDDS as middleware for autonomous vehicles. NDDS is also now installed in autonomous submarines that perform underwater service, installation, exploration, salvage and recovery operations. In addition, RTI's middleware is installed in Robonaut, a humanoid robot developed jointly by the Defense Advanced Research Projects Agency and the Robot Systems Technology Division at NASA's Johnson Space Center.

RTI provides NDDS middleware for microprocessors from companies such as ARM, STMicroelectronics and Infineon. ARM offers its software development tools separately, which the company says can be used to integrate processing and communication tasks for body, chassis, safety and powertrain systems.

"ARM cores are now running in 65% of ABS and chassis control systems worldwide," said Wayne Lyons, director of embedded systems at ARM Ltd. "Choosing ARM cores allows automotive design engineers to consolidate their architectures and reduce them from the 20 or more architectures they use today to a few."

Instead of maintaining software development tools for all these architectures, "you can standardize on our tools and focus on continuous innovation, eliminating all the problems that come with compiling from scratch for each new model," said Lyons.