Visit CMU: The birthplace of driverless technology

　　Because it always runs faster and farther than reality. Thousands of years before the invention of rockets, there were already records of space exploration. The same is true for unmanned vehicles. Let's follow the automotive electronics editor to learn about the relevant content.

　　In 1984, before GPS was invented, the Carnegie Mellon University team tested their first self-driving car in an abandoned parking lot. Today, the level at that time is not worth mentioning, but since the 1980s, with the funding and promotion of institutions such as DARPA in the United States, a series of research teams have explored the road from scratch to the development of driverless technology, which is not easy. This has also made a number of colleges and laboratories including Carnegie Mellon University, which is still the cradle of driverless technology and talents.

　　Today, driverless cars are closer to people's lives than ever before. So what kind of story has Carnegie Mellon University, which gave birth to the prototype of this technology, experienced? Leifeng.com (public account: Leifeng.com) and New Intelligent Driving compiled and translated this report from the Pittsburgh Business Times for readers.

　　In an abandoned parking lot in Pittsburgh, USA, an unmanned vehicle is driving slowly around the site. It has a name, called Terregator. The person who created it said, "It is like a rolling table that can identify and drive autonomously."

　　"Autonomous identification and driving" is the attribute that is often used to describe self-driving cars today, making this "rolling table" almost make history.

　　That year was 1984, long before GPS was invented. So, in order to track the driving trajectory of the car, the team tied a paint can with a hole in the bottom to the rear of the car. As the car moved, the leaking paint drew a path on the rear of the car.

　　When mentioning this incident, Red Whittaker, a professor of robotics at Carnegie Mellon University (CMU), still remembers it vividly. The prototype vehicle at that time was the result of CMU's years of research and development of the Terrestrial Navigator (Terregator). At the same time, the research team also spent several months designing the appearance and structure of the vehicle.

　　The car, named "Terregator" by the CMU research and development team, has six wheels and can travel several centimeters per second. A series of sensors on the body, including sonar rings, cameras, and a single-line lidar rangefinder, will be responsible for sensing obstacles and the environment.

　　“I don’t want to claim that this is the first in the history of driverless cars, but for the team that developed it at the time, Terregator holds such a place in their hearts,” Whittaker said with emotion. “The first time I saw it there, even though it was just circling the parking lot, the surprise and the sense of wonder were indescribable.”

　　Since then, CMU has become the birthplace of autonomous driving technology because of the birth of Terregator, which has also opened up CMU's more than 30 years of technical research on this basis. Over the years, Whittaker, the creator of Terregator, has also been supporting the research and development of various other automotive-related technologies.

　　Today, if a self-driving car appears on city streets, it will no longer be so surprising, but CMU and the city of Pittsburgh where it is located have become the oldest place to breed driverless technology because of the story of Terregator.

　　In fact, CMU's research and development of autonomous driving technology began with an opportunity in the mid-1980s.

　　At that time, the US DARPA (Defense Advanced Research Projects Agency) was operating a 10-year "Strategic Computing Program". DARPA hoped to benefit from the rapid development of computer architecture, software, and chip design, and to push AI technology to new heights.

　　In 1983, the U.S. Department of Defense listed Autonomous Land Vehicle (ALV) as one of the research projects of the Strategic Computing Program and formulated an annual plan.

　　1985: Road tracking test, the vehicle traveled at a speed of 10km/h on a paved road without obstacles.

　　1986: Obstacle avoidance test, the vehicle travels at a speed of 20km/h and can identify and avoid fixed obstacles.

　　1987: Off-road route planning experiment, planning a vehicle route and passing through open desert areas at a speed of 15km/h.

　　1988: Highway network route planning and obstacle avoidance test, planning vehicle routes, and achieving driving on the highway network at a speed of 20km/h with the help of road signs, as well as completing map correction and bypassing obstacles from the roadside.

　　During that time, DARPA funded a number of institutions and manufacturers, and as one of them, CMU was tasked with solving the complex perception and integration problems of the ALV system. In order to overcome this technology, CMU researchers established a navigation laboratory in 1984, named "NavLab", focusing on the research of difficult visual perception problems in complex environments.

　　"In that era, everything started from scratch. From planning paths for self-driving cars, to path tracking, to obstacle avoidance, to finding the right software system for automation. How do machines see the world? How do machines understand the world? Everything is too rudimentary today, and for this reason, the achievements at that time can be called real inventions," said Whittaker.

　　How do robots see the world?

　　Whittaker will never forget the day in 1984 at Schenley Park in the US when the team decided to test the Terregator’s real-world obstacle avoidance capabilities for the first time.

　　“It was late spring in Pittsburgh, and all the universities were still on vacation, so many students were out walking and enjoying the sun,” he said. “So, we tested the Terregator as they all gathered in the park. It slowly drove alongside these students, and it also moved around some of them who were lying on the park lawn. The students were seeing the Terregator for the first time, just as the Terregator was seeing them for the first time. It was a really cool moment.”

　　However, the challenge for autonomous driving is not just the obstacles themselves, which means that the vehicle needs to perceive these obstacles, and more importantly, decide how to drive next.

　　"There is a lot of work to be done on the driving decisions of self-driving cars. In other words, once the visual data is interpreted, the vehicle must make judgments on how to drive, how to turn, etc.," said Martial Hebert, director of the CMU Robotics Institute. "I think this is where innovative technologies burst out."

　　Hebert recalled that when CMU first started developing self-driving car technology, their car only successfully "saw" for two days. At that time, the team used camera-like sensors, combined with image analysis technology, and used lidar to directly obtain 3D perception information. These two methods of environmental perception are still used today, but they have become more complex than in the 1980s.

　　The most recent self-driving car developed by CMU was a Cadillac SRX that debuted in 2013, successfully carrying then-U.S. Pennsylvania Senator Bill Schuster to the airport.

　　Technically, the car uses image analysis and measurements of road signs such as lane lines to assist in environmental perception, and the car is already able to achieve average highway driving speeds. What really distinguishes it from the Terregator is not only its speed and appearance, but also its ability to perceive dynamic objects in the environment, whether it is a person, a car, or a bicycle.

*Autonomous driving modified car Cadillac SRX

　　“When you have to deal with the relationship between the self-driving car and other participants in the traffic environment, such as pedestrians, things become very complicated,” Hebert said. “The vehicle must make driving decisions in a completely dynamic traffic environment.”

　　In modern research, the driving decisions of vehicles will become more detailed. Hebert recalled that after Terregator, in 1986, CMU also developed an autonomous vehicle based on a modified Chevrolet van, named Navlab 1, which could reach a speed of 20 miles per hour (about 32 kilometers per hour).

*Autonomous driving modified car NavLab 1

　　After Navlab 1, CMU began to focus its research and development on more detailed autonomous vehicle behavior decisions, that is, to make the vehicle "behave more like a human driver." This means that autonomous vehicles are not just about "avoiding obstacles and moving forward," but need to combine more scenarios and analysis to make judgments.

　　“So the next step was to enable the vehicle to infer the movement of other vehicles, such as a car passing by,” Hebert said. “That was a critical advance because it allowed the self-driving car to move from simply sensing and tracking the surrounding environment to dynamically reacting to the vehicles around it.”

　　These research results were all demonstrated in the NavLab 5 subsequently developed by CMU, an autonomous driving car modified from a Pontiac sports sedan. In 1995, the car drove from Pittsburgh, Pennsylvania to San Diego, California, a distance of 4,587 kilometers in the "NO Hands Across America" ​​(NOAA) experiment, of which the autonomous driving part reached 98.2%.

* Autonomous driving modified car NavLab 5

　　Todd Jochem, a CMU doctoral student and one of the R&D members of Navlab, recalled that it took them four months to complete the vehicle modification and software debugging, with a total cost of no more than $20,000. All the equipment included a computer, a color camera, GPS, and a fiber optic gyroscope.

　　Interestingly, their GPS was not used for positioning, but for speed measurement. Todd Jochem said that at that time, GPS had not yet opened the high-precision positioning function, which means that the price of using this service would be very high. At the same time, even if GPS was used for high-precision positioning, they did not have a matching map.

　　After that, in 2005, CMU participated in the DARPA challenge with H1ghlander, a modified General Motors Hummer that successfully overtook a human-driven vehicle on the road.

　　“I still remember the excitement of seeing a self-driving car overtake a human-driven car,” Whittaker said. “Now that everyone knows how it was done, I don’t want to diminish the value of what we accomplished.”

　　on the way

　　In the 2007 DARPA Urban Challenge, 11 self-driving cars stood out among more than 100 participating teams due to their outstanding performance. Among them was Boss, CMU's self-driving modified car based on the Chevrolet Tahoe. In the end, it won the championship in that year's competition with an average speed of 22.53km/h.

　　The predecessor of the DARPA Urban Challenge was the "DARPA Grand Challenge", which is equivalent to the Olympics in the driverless circle. In 2007, the challenge was held at a retired Air Force base in the United States, a two-hour drive from Los Angeles, and the "DARPA Grand Challenge" was also called the "Urban Challenge". At that time, the track was 55 miles (about 89 kilometers), and the participants had to complete three driving tasks during the process and finish the race within 6 hours.

　　"When the driverless car started the first of three driving tasks, the boss suddenly stopped moving, and everyone's heartbeat seemed to skip a beat," said CM who participated in the challenge.

　　"When the driverless car started the first of three driving tasks, Boss suddenly stopped moving, and everyone's heart skipped a beat," recalled Raj Rajkumar, a professor in the Department of Electrical and Computer Engineering at CMU who participated in the challenge.

　　"We were supposed to be the first to start, and when we were about to start the system, the Boss stopped moving!" Rajkumar said. "Everyone in the team was just like, 'What the hell is going on?'"

　　After a brief investigation, the team found the problem. Unlike other participating cars, Boss was very close to a huge display screen reporting the race situation at the start of the track, which interfered with Boss's GPS. Afterwards, Boss continued to participate in the race and won the championship with a lead of 20 minutes over the second place.

　　CMU's outstanding performance in the DARPA challenge directly prompted General Motors to donate US$5 million to establish a second laboratory, the Autonomous Driving Technology Collaborative Research and Development Laboratory, with Rajkumar as co-director.

　　Rajkumar is also the founder of Ottomatika, an autonomous driving technology startup. In 2013, Ottomatika was founded and is committed to commercializing the results of autonomous driving research developed by CMU. Subsequently, the company cooperated with Delphi to develop some active safety platforms for cars driving in cities or highways.

　　In 2015, Ottomatika was acquired by Delphi. In the same year, Delphi used Ottomatika's technology to modify an Audi Q5 self-driving car, driving from San Francisco to New York, with autonomous driving time reaching 99%.

　　To date, CMU has published 140 research results on autonomous driving technology, among which Cadillac SRX has become a highlight, not only because of its pioneering deep integration of many cutting-edge technologies, but also because many sensors and kits have been embedded and integrated with the vehicle.

　　"Boss has a bunch of equipment on its body, including a lidar on the roof, a bunch of cameras, and a lot of electronic devices at the rear. There are two huge display screens inside the car, and the driver needs to constantly monitor this information. At the same time, even experienced drivers with 30 or 40 years of driving experience must be trained to know how to operate this self-driving car," Rajkumar described. "In 2011, we worked with General Motors to modify the Cadillac. The primary goal at the time was to make the car look more normal."

　　In the Level 0-Level 5 autonomous driving levels defined by SAE, Rajkumar believes that SPX has reached Level 3.3. As for reaching the true Level 5, Rajkumar believes that it will take at least 10 years, or even longer.

　　"Next, we need to face complex problems such as weather conditions, tunnels, and bridges," Rajkumar said. "In tunnels and bridges, GPS fails. We can currently achieve 85% autonomous driving, but the remaining 15% must always be thoroughly solved."

　　"Now, we are preparing for the realization of the next generation of autonomous driving. This is no longer just about autonomous driving itself, but also involves the development of many automotive-grade components. In order to truly realize productization, we must also consider the complex issues of various urban roads and weather issues. Of course, we must ultimately control costs."

    The above is an introduction to Automotive Electronics - Visit CMU: The Birthplace of Driverless Technology. If you want to know more related information, please pay more attention to eeworld. eeworld Electronic Engineering will provide you with more complete, detailed and updated information.