The Academic Motorsports Club Zurich (AMZ) was founded in 2006, and every year since then, the club has produced a prototype to compete in Formula Student competitions all over Europe. It started by building combustion-powered cars and then began developing fully-electric racing cars in 2010. Since the introduction of the Formula Student Driverless class in 2017, AMZ has become a leader in the development of autonomous race cars.
Victor Reijgwart is AMZ Driverless’ Head of Estimation. As he told Velocity Magazine: “Participating in Formula Student Driverless is an amazing opportunity to put the knowledge of robotics we acquire in courses into practice. This is so exciting because the racing aspect of the competition challenges us to constantly push the performance envelope by trying new algorithms and system architectures.”
Formula Student Driverless teams compete in five dynamic and three static disciplines. The dynamic disciplines each feature a unique track layout, presenting different challenges for vehicles and their autonomous systems. The most challenging are Autocross and Trackdrive, where the cars race at high speeds on an unfamiliar track shaped like a small F1 circuit. During the static disciplines, the teams defend their designs before a panel of experts from academia and the automotive industry.
Reijgwart said the AMZ project is using products from Hexagon | NovAtel to help with two specific tasks. “ First, we need to be able to estimate the velocity, position and heading of the car. Having good estimates is crucial, allowing us to drive reliably and fast. The second thing is we have to create ground truth maps of our test tracks. These precise maps, together with accurate knowledge of the car's position, give us a bird’s -eye view of how our autonomous car performs when testing.” Reijgwart said the quantitative evaluation of the different parts of the system enabled by NovAtel technologies helps the team to focus its efforts on the true bottlenecks, which were much harder to pinpoint in the past. “Better insights also help us to develop the system faster and configure it better,” he said, “despite its growing complexity.”
The AMZ team itself, Reijgwart said, is made up of robotics students who did not previously have extensive knowledge in the area of GNSS. “But while reading up on the various products that were available, it became clear that NovAtel offered high-performance products that also met the ruggedness level we needed in order to use them in a race car.”
Key Enabling Components
AMZ is using multiple products from NovAtel. Reijgwart said the most important one is the PwrPak7D-E1 ™ with dual GNSS-303L antennas, serving as the car’s Inertial Navigation System (INS). “ Fusing data from the PwrPak7D-E1 with wheel odometry measurements gives us a high-rate estimate of the car’s motion. he said. “This motion estimate is directly used by the car’s control, localization, mapping and environment perception systems. Since it plays such a critical role, it was important to us to use a reliable product from a trusted brand. Furthermore, the PwrPak7D-E1 was the only sensor that combined Dual-GNSS, a high-quality IMU, RTK and CAN interfaces within a single , small product.”
Reijgwart also highlighted the importance of the dual antenna: “This way we can measure the lateral and longitudinal velocity components, as opposed to single antenna setups that can only measure the velocity’s magnitude. Having better estimates of the car’s lateral velocity directly improves the performance of our control systems. Furthermore, the absolute heading measurement from the dual antennas helps us create better maps. Finally, RTK corrections make the position and heading measurements accurate enough to be used as ground truth. This is allowing us, for the first time, to validate our system’s performance quantitatively. Together with NovAtel, Leica was kind enough to provide access to their HxGN SmartNet. This makes it possible to measure ground truth car poses.”
The AMZ race car also includes another key NovAtel system, as Reijgwart explained, “We also have the PwrPak7 ® with a GPS-702-GG antenna and SmartNet RTK corrections. This serves as our ground truth mapping device. In the past, we mainly relied on qualitative evaluation of the car’s behavior. But given the steady increase of our autonomous system’s complexity over the past two years, it has become indispensable to have quantitative, repeatable and detailed insights into the system’s performance. Using these sensors, we created a handheld device that lets us create ground truth maps of our race tracks with two-centimeter accuracy. Formula Student race track boundaries are solely marked by cones, so the mapping process consists of walking around the track and recording the pose of each cone. Combining this ground truth track knowledge with ground truth car poses allows us to validate each part of our autonomous racing pipeline, from perception to estimation, mapping and control.”
The PwrPak7, with a survey-grade antenna, Reijgwart said, in combination with the PwrPak7D-E1 on the car, “give us complete knowledge of the state of the vehicle and the environment within the same frame of reference. When we're going into a competition, it is not allowed to map the race track in advance. But having ground truth while testing makes it possible to directly measure the errors in the vehicle’s onboard environment perception, mapping, localization, planning, motion estimation and motion control systems. This, in turn, makes it possible to quantitatively compare different algorithms.”
All of these quantitative metrics can be calculated without human intervention, Reijgwart said, and they can be used to extend the existing data management platform with key performance indicators that track the progress of each subsystem’s development over time. “Another thing we intend to do is to validate each autonomous subsystem individually. For instance, we previously weren’t able to test how our planning and control algorithms would perform if they received a perfect map and pose estimate. Knowing this now allows us to focus our efforts on improving the parts that matter most.”
Finally, he explained, ground truth will make it possible to measure how high-level factors such as the perception range, delays, planning horizon and so forth affect system performance. “In the racing industry, this is commonly referred to as a sensitivity analysis. With NovAtel’s system , we will be able to conduct this analysis for the first time , and its results will shape our fundamental design decisions for our future cars.”
Racing World Meets Real World
The Formula Student competitions involve technology research and development, but Reijgwart said he does see some potential realmarket applications. “What sets driverless race cars apart from general autonomous driving,” he said, “is their primary focus on controlling vehicles at their physical limits. Just like in traditional racing, we aim to minimize lap time by reliably driving around the track at the highest possible speed.”
“Racing without a driver,” he said, “introduces many new challenges, from perception of the track boundaries to estimation and control of the vehicle’s motion. In addition to delivering an exciting experience, autonomous racing provides a safe, controlled environment for testing and developing technologies that will eventually benefit many industrial and consumer products.
“Potential applications range from improved reliability for service robots to autonomous vehicles. Many parallels can be drawn between the challenges in racing and emergency scenarios encountered by cars on public roads, including high-speed avoidance maneuvers and safe driving despite degraded road conditions.”
Reijgwart said he believes some of the fundamental components used for autonomous racing already overlap with technologies being deployed in current Advanced Driver Assistance Systems (ADAS). “Hopefully,” he said, “more advanced features, such as collision avoidance maneuvers that go beyond preventive braking, will be deployed soon. In the long term, autonomous racing research findings could play a big role in the development of fully autonomous (SAE level 5) vehicles that are able to safely transport passengers under all circumstances and without any human input.”
Satisfying Design Requirements
The collaborative process between AMZ/ETH and NovAtel has been positive and rewarding, Reijgwart said. “The first contact took place in early December 2018, when I spoke
to [vice president of sales] Steve Duncombe over the phone about our interest in working with them. I called them after we finished combing through existing GNSS products, carefully comparing them, and analyzing how they would fit our design requirements. NovAtel’s GNSS products fit very well within our overall concept, and Steve promptly put us in contact with [business development manager, safety critical systems] Andreas Niemann, with whom we discussed our goals and what sensor choice would be most suitable.
“NovAtel was so kind as to sponsor us and allow us to use their products, and furthermore arranged sponsorship from Leica so that we could use HxGN SmartNet. Once the products arrived, NovAtel application engineers gave us feedback on our sensor placement and helped to configure the receivers to work well with the rest of our system.” Reijgwart said, “The collaborative process was great, and their products lived up to the high expectations we had. So we would certainly be happy to keep working together in the future.”
AMZ has secured the overall first place at each of the Formula Student Driverless competitions since they began in 2017. “In addition to taking part in the competition, we
open-source parts of our code base and regularly publish our research findings,” Reijgwart said. Anyone who happens to attend this year’s Student Formula events will have no trouble spotting the AMZ car. The NovAtel logo will feature prominently, on the track and, fingers crossed, on the awards podium.