Algorithms race to save lives

Close up view of green Hexagon branded Indy car

It is generally—though not universally—acknowledged that the autonomous, driverless environment will, if fully and properly achieved, be safer than one populated by error-prone humans. That’s why departments of transportation are eager to move in that direction. There are a few bumps in the road, however. Note the words “fully and properly achieved.”

Numerous factors will have to be properly accounted for, fed into artificial intelligence (AI) engines and supplied with highly complex, highly integrated data streams for that safe driverless environment to become a reality.

Initiatives are underway around the world to make this happen. Some of the youngest and brightest minds engaged in the effort will meet bumper-to-bumper in the Indy Autonomous Challenge (IAC). This $1.5 million competition carries a $1 million grand prize for programming driverless racecars in the world’s first high-speed autonomous race at the Indianapolis Motor Speedway.

Thirty-one competing teams from universities around the world prequalified for the event. By the time actual driverless racecars wheel up to the IAC starting line in October 2021, the field will have narrowed to 10 through a series of qualifying rounds, some in simulation, some as individual trials on the famous Indy Oval. All competing teams get the same vehicle to work with, a Dallara AV-21, modified to carry no one in the cockpit.

In learning how to make driverless cars perform safely at hyper-highway speeds amid frenetic lane- and lead-changes, IAC participants will gain a vast storehouse of powerful knowledge. In their coming careers, they will help enable safe autonomous operation in city traffic and during freeway commutes. The student-researchers competing in the IAC will carry that experience into the auto industry, inaugurating the next giant steps in driverless capabilities.

“There are real-world problems we are solving with this competition,” said Matt Peak, managing director of Energy Systems Network, which organized the IAC in partnership with the Indianapolis Motor Speedway. “Saving human lives. If these students can race and maneuver using commercially available sensors on these racecars, then we can equip vehicles on the road to do that, to avoid other vehicles, things we don’t see, things we aren’t fast enough to handle. The ultimate goal is to save lives.”

The IAC had its genesis in the Defense Advanced Research Projects Agency’s (DARPA) Urban Challenge, the third of three multimillion-dollar prize events (2004, 2005, 2007). The DARPA Challenges created a mindset and a research community that now, a decade-and- a-half later, has produced fleets of autonomous cars, shuttles and other ground vehicles. Some of the organizers and sponsors of the IAC cut their teeth on the DARPA Challenges. Now, they are passing the baton, encouraging a younger generation to take the next lap.

The IAC introduces a few new wrinkles, keeping pace with the rapid development of many contributing technologies since 2007. Additional factors have been placed in the mix to better suit the outcomes to production-model driverless cars.

For one, all teams will race with identical vehicles, powertrains, sensor systems, drive-by-wire systems and processors. The only variable? Their autonomy-enabling software stacks. “This race will go to whoever has the best software algorithms and neural networks,” said Lee Baldwin, segment director, core autonomy at Hexagon’s Autonomy & Positioning division. Hexagon is a vehicle sponsor of the IAC, providing IAC competitors with positioning hardware, software and expertise from Hexagon’s Autonomy & Positioning division’s leading brands, NovAtel and AutonomouStuff.

“The Dallara-built IAC racecar is the most advanced, fastest autonomous vehicle ever developed,” said Paul Mitchell, president and CEO of Energy Systems Network. “IAC sponsors are providing GNSS receivers, radar, LiDAR, optical cameras and advanced computers, bringing the value of each vehicle to $1 million.”

To acquire one of the 10 IAC Dallara AV-21 racecars to compete in the final rounds, university teams needed to pass a simulation qualifying round, the first critical milestone in narrowing the field from the initial 31 prequalified entrants. That’s where the magic—the software—must happen. “For the Indy cars we don’t install the software stack; that’s up to the teams,” said Baldwin of the work that Hexagon | AutonomouStuff contributes.

“They have to figure out their own perception system. Don’t worry about stoplights but do worry about interaction with other vehicles. The Dallara racecars have been overspec’d with sensors, cameras, LiDAR, radars. The teams have a superset to use, for all the high-speed use cases. Autoware wouldn’t fuse that many sensors. The teams will have to write their own software to do all that.”

Powered by GNSS

Hexagon’s Autonomy & Positioning division provided GNSS receivers and subject matter experts to Clemson University’s Deep Orange 12 team of highly skilled graduate students and professors. Clemson is not competing in the race but has architected the overspec’d sensor kit for the Dallara reference vehicle, which AutonomouStuff will then duplicate times 10.

“We have two GNSS units,” said Chris Paredis, professor, BMW Endowed Chair in Systems Integration at Clemson University, and director of the Deep Orange program. The latter is a vehicle prototype master’s degree program offered annually by the Clemson University International Center for Automotive Research.

“The two receivers are PwrPak7-Ds with full multi-frequency, multi-constellation satellite configuration.” The PwrPak7-E1 contains a micro electromechanical system (MEMS) inertial measurement unit (IMU) to deliver Hexagon | NovAtel’s SPAN technology, a deeply coupled GNSS+ inertial engine in an integrated, single-box solution. It has a powerful OEM7 GNSS engine, integrated MEMS IMU, built-in Wi-Fi, onboard NTRIP client and server support, and 16 GB of internal storage. The PwrPak7-E1 also has enhanced connection options including serial, USB, CAN and Ethernet.

“The important part,” Paredis continued, “is we have two antennas attached to each GNSS unit. One has an antenna on the left-side pod and one on the right-side pod, the other has one antenna on the front of the cockpit and one on the roll hoop. That gives heading and that’s quite important to know in exactly which direction the vehicles are pointed.

“We did quite a bit of testing at the IMS Track and Lucas Oil Raceway. The GNSS reliability and accuracy is just phenomenal. We are relying on the NS PVA logs; they give us information at 100Hz on both GPS and INS units.

“I expect the teams will rely really heavily on the GNSS data to navigate the track.”

The Deep Orange 12 team used HxGN SmartNet RTK corrections, which brought the accuracy down to a few centimetres.

As for integration with the rest of the system, each vehicle has primary compute node contributed by sponsor ADLINK, an Intel Xeon processor with a high-end general-purpose GPU. The teams will use it for the perception pipelines and fashioning their highline behavioral planner. That computer is connected to a high-speed switch to which all the peripherals are connected. It has a capacity of 40 gigabits per system for the six cameras, three LiDARs and the two GNSS. Each Dallara will also carry four radars, directly connected to the computer, because their interface is CAN-based rather than Ethernet-based.

For communication purposes, going through that same switch is a V2X wireless communication system by Cisco Ultra-Reliable Wireless Backhaul, communicating with zero-latency handoffs between base stations at high speed. Cisco Systems will install the wireless base stations on the track for the IAC in October, with very directional antennas.

“If you were to rely just on Wi-Fi,” Paredis explained, “as the racecar moves from one access point to another, there is typically a short interruption in the service, which can be up to a second or so. With the cars covering 80 metres per second, that much interruption is unacceptable.”

The system will be set up with a “make before break.” That is, make contact with the next base station before breaking contact with the previous station.

The onboard computer will communicate with the drive-by-wire layer on the racecar: three electronic control units (ECUs) for breaking+steering, powertrain, and power distribution. These are connected to an overarching ECU, the Raptor, by New Eagle.

Deep Orange 12

Clemson’s Deep Orange 12 design team conducted a weekly town hall meeting to share and discuss details of the vehicle as it was assembled, working hard to have a good understanding of what it’s capable of, and how to interface with the vehicle.

Rob Prucka, as associate professor at Clemson, heads Deep Orange 12, designating the 12th year Clemson has run this graduate vehicle prototyping program that builds from custom requirements, although this is only the second autonomous program. Next year’s Deep Orange 13 will be autonomous, he asserted.

“We communicate weekly with 400 students around the world,” Prucka said. “We give them updates on the vehicle, ask for their input preferences. It’s a pretty fruitful relationship with those teams. Obviously, we can’t give anyone a competitive advantage, so we communicate equally and to all.

“There are 30 different industry partners on all aspects of the vehicle that we’re working with, supplying pieces, all of them really critical within the vehicle. Our Clemson grad students are trying to manage the integration of all these partners, and the teams.” It’s a great management experience as well as an engineering design one.

The factory

The reference vehicle of the autonomous Dallara AV-21 headed to AutonomouStuff headquarters in Morton, Illinois, to be reproduced. All the sensors were integrated, wiring harnesses installed, computers and more. “We’re the factory,” Baldwin said.

AutonomouStuff had been in regular contact with Deep Orange 12, helping the team sort through various issues and decision trees.

The company took a large role in the specification of all the components because it has built more autonomous vehicles used for research than any organization in the world. For racing at high speeds, a key calculation was the required object-detection distance, based on the system’s response speed and stopping distance. This made it necessary to get sensors that had the longest range possible. Because the Indy track is a controlled environment, that simplified matters to some extent. For a standard vehicle on the highway, designers have to account for many unknowns, which is one of the reasons autonomy is such a complex concept to develop and deploy in a real-world environment. 

Further, all components except the computer had to be commercial-off-the-shelf, available on the market. No sensors could be custom-made. The IAC is pushing the limits, which of course is its purpose.

AutonomouStuff has a unique perspective, with a broad understanding of countless applications, thousands of successful technical demonstrations and customers located all over the world. That positioned the company to become a guiding light of the IAC.

Qualifier events

Several pre-race events demonstrated development prowess in a range of environments to narrow the competing field from 31 teams to its 10 contestants. “There are performance metrics that the teams have to satisfy in order to get on the track,” Peak said. “A minimum threshold they have to cross.” 

Hackathons gave the university teams certain problems to solve. These established whether the teams were committed to the overall race and could work through the challenges. As they progressed through the qualifying rounds, they had to demonstrate they could take their software and make it perform to certain standards, qualifying to get an actual racecar and track time.

A mapping vehicle collects highly precise data on the track, and the teams import that data into their software, replicating the environment to test out their algorithms against what they’re about to encounter.

When the vehicle model is put into the software, it’s important to enterall the physics: the point of slip, speed, weight, grip pattern, tread, g-forces and a number of other parametres. Once the software is deployed, if the physics are done right, the racecar should respond

The successful semifinalists got their very own brand spanking new Dallara Av-21s. The first one was unveiled in May at the granddaddy of them all, the Indy 500 itself. Some teams also got to attend the legendary race. That’s when they made their first acquaintance with the Oval and the outfitted autonomous Dallaras.

The teams, from Germany and Italy to India, all the way round the globe to South Korea and Hawaii, and back to the usual suspects in the tech university sphere in North America, each got their collective minds behind their respective vehicles’ driverless wheels. At various times during the coming months, they’ll get their own track time to actually run on the Speedway.

The finish line

The teams have miles to go before they sleep—if they sleep at all before the final race in October.

That’s not the end of the story, though. It goes on with the lessons learned on the racetrack, with the students equipped with that priceless experience and the knowledge of how to manage safe autonomous driving. They will carry that forward with them into their auto industry careers.

Half the motivation for the IAC was getting cars to perform. The other half, more than half, was getting college students hands-on experience with autonomy so they’re better prepared to enter the industry.

AutonomouStuff and many other companies, from the big car manufacturers themselves to Tier One suppliers
and more, will be looking out for these team members, eager to welcome them into the expanding autonomous work force.

Another motivation for the race that IAC CEO Paul Mitchell mentioned was to get the general public more comfortable with autonomous cars. That is to say, if a racecar can run safely and autonomously at high speed, then maybe driverless vehicles can do so on a public road some day.

The learning curve is very steep. The payoff will be far more than worth it.

Read the full PDF here: