Audi's autonomous Audi TT conquers Pikes Peak - how long before it betters a human driver?
By Mike Hanlon
November 19, 2010
Not long ago, there was informed debate on whether a purpose-built computer would ever beat a chess master. Nowaday, a HTC mobile phone can achieve Grand Master status. Computers continue to get exponentially faster, not to mention considerably smarter through improved software, whereas humans are effectively nearing their limits. Hence, it’s arguably only a matter of time and R&D; focus before computers (plus improved sensors and software) surpass any specific human capability. This week Audi revealed that its Autonomous TTS research car had completed the 12.42-mile Pike’s Peak mountain course in 27 minutes. An expert driver in the same car would take around 17 minutes – now we have a benchmark, the race is on, and it's almost inevitable that a computer will one day outdrive the best of our species, and it may be sooner than you think.
Humans are not very good at driving cars, as is evidenced by our ability to destroy 1.3 million souls on our roads each year. Our deficiencies for the task of safely controlling a car on public throughfares are many. We are almost incapable of driving safely while multitasking, are ridiculously easily distracted, take all manner of mind-altering substances before we drive, continually take imprudent risks and our situational awareness is largely restricted to our field of vision, which is but a sliver of the ideal 360 degrees.
By comparison, computers can monitor a full 360 degrees plus thousands of variables simultaneously, are diligent and attentive in the extreme, and calculate each and every risk, erring on the side of caution to exactly the degree to which they are programmed.
That’s why Toyota, General Motors and Volkswagen are spending so much money in the field of autonomous vehicles – autonomous vehicle expertise is already being used to make us safer in our cars and over the next few decades, the deployment of more and better intelligent automotive systems will help to stop us committing genocide on such a grand scale.
Hence the ascending of the demanding Pikes Peak mountain course by Audi’s autonomous TTS in September is a significant achievement in that it sets the benchmark, visible for the first time, as to how close autonomous vehicle are to the best human drivers.
It might seem like a lot of difference in the times between an expert driver and a computer in the TTS (17 minutes compared to 27 minutes), but I must confess to being staggered that the autonomous vehicle was immediately so close to the theoretical best – roughly an extra 59% over the fastest time of the best human driver. In Grand Prix racing, where the best drivers and riders in the world compete, lap time differentials of 7.5% are allowed by regulation, and quite often greater differences are tolerated. The difference in lap times between an expert driver and an average commuter would also be dramatically different, so the Audi TTS might already be much closer to the average person's driving ability than the figures suggest.
The 265 bhp research car has been cooperatively developed by Volkswagen/Audi in conjunction with Stanford University and Oracle, and it is unquestionably progressing at a phenomenal rate. The 27 minute time was achieved the first time the vehicle achieved the 14,110-foot summit in Colorado without stopping. Five other times during the week-long testing, the car ran the complete course, pausing briefly to verify route data. It is most likely to get much faster very quickly.
Admittedly, it's still a long way from cracking the ten minute mark being approached by Monster Tajima in his 900 bhp Suzuki, but a significant start has been made and the approach has been far more cautious than the way humans tackle timed courses. It's an old racing adage that it's a lot easier to make a fast driver who crashes safe than to make a slow driver faster. The penalty for error on Pikes Peak is massive as the edge of the circuit is often a massive cliff. Audi is logically taking a cautious and considered approach because the negative publicity of a car plunging over a fatal drop would hinder the development.
Volkswagen’s Electronic Research Lab and Stanford have been at the forefront of autonomous driving research for several years, having won the second DARPA Grand Challenge and finishing second in the most recent DARPA Urban Challenge held in Victorville, California in 2007.
From the Audi press statement: The Autonomous Audi TTS Pikes Peak reflects the ERL-Stanford strategy of conducting research in tiers that thoroughly investigate technologies needed to perform different autonomous driving tasks ranging from low-speed maneuvering in urban environments to high-speed handling on varied road surfaces on a challenging course like Pikes Peak.
When research began on the Autonomous Audi TTS Pikes Peak research project just over a year ago, the direction was clear: employ emerging software, algorithms and electronics to help everyday drivers stay in control, and safely on the road, even during extreme driving conditions.
The aim of the Autonomous Audi TTS Pikes Peak research was to develop a technology that would enhance a driver's abilities, much as computerized systems of passenger jetliners assist skilled pilots.
"We are not trying to replace the driver," said Professor Chris Gerdes of Stanford University, "Instead we want to learn how the best drivers control the car so we can develop systems that assist our robotic driver and, eventually, you and me." Working together, Audi, Stanford University, the Volkswagen Group Electronics Research Lab and Oracle developed a distinct engineering achievement. The Autonomous Audi TTS Pikes Peak integrates advanced algorithms, the Oracle Java real-Time System (Java RTS), Oracle Solaris and GPS with safety and navigation systems found in stock Audi TTS models to maintain control at a physical performance extreme.
Java and Oracle Solaris provide a significant advancement over traditional execution models in terms of reliability, transparency, debugging capability, programming model, predictable response-time characteristics, and cost. Using the standard Java programming model and memory management functionality, developers were able to program the Autonomous Audi TTS to easily differentiate processes based on their importance and precisely determine when time-critical functions should be executed.
"Oracle Java RTS is the first enterprise-class solution to formally address the issues of latency and unpredictable response times for Java applications and we're thrilled to be part of this research project," said Greg Bollella, chief architect, Embedded Java, Oracle. "For the Autonomous Audi TTS Pikes Peak, Java was used to acquire GPS position coordinates and distribute those coordinates to all of the other components in the system. It also served as the safety controller for the vehicle, responsible for gracefully bringing the car to a stop if any of the traditional systems malfunctioned."
Deciding on a location to prove the technology was an easy choice as the Pikes Peak route offered steep inclines, switchbacks and varied road surfaces for the autonomous Audi TTS to navigate. Pikes Peak is also the place where Audi technology became legendary in the rally racing world a generation ago thanks in large part to the distinctive quattro® technology, which this year celebrates its 30th anniversary as a technological all-wheel-drive breakthrough exclusively on Audi vehicles.
The Autonomous Audi TTS Pikes Peak achieved this year's goal on the mountain. The next stage of the research project will involve autonomous high-speed handling on paved surfaces. The research team is evaluating race tracks where they can conduct the next phase of this research.
In keeping with all trials of the technology, the Autonomous Audi TTS Pikes Peak research team worked with local authorities to conduct the mountain testing during closed-course runs that emphasized public safety.
To celebrate its accomplishment, Audi is featuring the Autonomous Audi TTS Pikes Peak at its inaugural display at the SEMA Show, which begins today in Las Vegas. ENDS
This is just the start. It’s an enormously significant event and a milestone on the road to much safer automobiles. Hats off to Audi, Stanford and Oracle for the achievement. We look forward to watching the progress and wonder not just how long it will take for a suitably equipped car to be able to best a Sebastien Vettel or Lewis Hamilton, but how long before drivers are judged by how close they can get to the "ideal time" set by a computer.
Indeed, it's a fascinating mental exercise to imagine the future of racing altogether - do computers have a place in motorsport? We're increasingly seeing F1 teams such as Ferrari employ simulators to hone their drivers and to test changes - it's not such a big step from advanced driving aids to fully autonomous any more.