How the Top Gear record happened and what it means for the future of autonomous transport
If it seems like sending an unmanned, electric heavy transport vehicle around a famous TV racetrack at record speed was just for fun… well, you’re not entirely wrong. Myself and the Einride team spent months preparing and over a month quarantined in a farmhouse in Surrey, and we would do it again in a heartbeat for the thrill of seeing our hard work result in a perfect racing line around Hammerhead, or a lap of the Goodwood circuit among some of the world’s fastest and most expensive vehicles.
But perhaps more important than the satisfaction of our achievement is what we learned about high-speed autonomous transport operation, and what it means for the future of self-driving electric freight. In the same way racing programs have resulted in countless automotive innovations we take for granted every day, our experience in the U.K. has implications that stretch far beyond the track.
How we made it happen
Making a safety analysis and case
It nearly goes without saying, but in autonomous vehicle operation, safety is always the number one priority. Whether in a closed facility, on a public road, or at a sparsely-used airfield, we perform an in-depth analysis at the start of every project to ensure safe operation, especially for a heavy goods vehicle at high speed. This analysis allows us to implement monitors and fallback procedures at the required integrity level to perform safe racing, the difference between executing successful laps autonomously, or ending up in a wall unexpectedly.
While professional racing drivers are known for their superhuman reaction times and for taking any risk necessary to gain an edge, racing a Pod is a much different endeavor. It involves engineering risk out of the equation, not the other way around. As a result, we do not rely on partners for the Pod’s safety architecture. Instead, our in-house autonomous drive development and safety components give us the possibility to balance the need to realize the optimal safety case with efficiency in implementation.
Perception and mapping approach
If an autonomous vehicle is going to navigate a racetrack — or nearly any drivable situation — it’s necessary to use the proper mapping and perception tools to make it possible. In the case of the two tracks in England, a lane-level static mapping procedure was sufficient. This gives us the ability to map the tracks quickly and efficiently, yielding more time to make the best possible lap happen.
With higher speed operation, there’s also a need for longer-range perception. Keep your eyes down the road ahead, not at the front of the vehicle (as you probably learned in driving school way back when…). If the Pod can’t detect what is farther ahead of it than just a couple dozen meters, it would not be able to reach the speeds necessary to complete the lap. As such, the Pod is capable of over 300 m of forward range, as well as advanced driveable surface detection, a necessity when racing on a flat airfield with little runoff between asphalt and grass in some places.
Seeking the optimal racing line
An example of how the Pod uses mapping, localization, and long-range lidar systems to find and adjust the perfect racing line.
In low-speed situations, finding the perfect line at which to take a corner is almost a non-factor. As long as an autonomous vehicle completes the turn safely and within the boundaries of the road, shaving centimeters or milliseconds off a cornering maneuver is overkill. At high speeds however — especially on a race track — even a self-driving truck like the Pod’s ability to clip the apex and follow through smoothly is crucial. There may not be many future applications for the Pod to go racing, but placing itself properly on a long, sweeping high-speed curve on a freeway will certainly come up.
To find the optimal racing line, the Pod localizes itself using multiple onboard sensors allowing it to position itself relative to the map of the track that it has been given. Then, based on the Pod’s position on the track, real-time optimization of the reference trajectory results in a perfect race line and the calculation of the steering and motor torques necessary to follow that line at the maximum possible speed. All the while, the Pod makes small adjustments for what can’t be known in advance, the same way you make tiny adjustments to your steering angle when taking a long curve, rather than holding the wheel in exactly one place.
Integrating planning, control, and monitoring
One of the most important factors in pulling off an achievement like the Top Gear and Goodwood laps is developing harmony between components from our trusted partners and our own proprietary autonomous driving architecture. Sending the Pod around a track at high speed requires synergies between Einride and our partners that stretch beyond a transactional relationship, but most importantly it builds the foundation for the scalability of these relationships.
Our understanding of vehicle dynamics and control together with our full grasp and ownership of the overall system architecture allows us to build simple and understandable interfaces with our partners, yielding a better safety case in critical control systems. Autonomous transport innovation isn’t a singular achievement, and will need to be a team off the racetrack and on the world’s roads as well.
What it means for AET
The next-generation Pod — the one you’ll be seeing on roads all over the world in the coming years — is designed for Autonomous Electric Transport (AET) levels 1 to 4 operation in commercial and public settings. That means we’re taking the use case of the Pod from testing and closed facilities to real-world, high-speed scenarios in countless different environments.
Everything we learned from the Top Gear record and Goodwood lap — and all that went into planning and executing it — is directly applicable to the development of high-speed autonomous operation on the road. Race-oriented safety analysis and procedures can be directly translated to highway operation and perception and mapping learnings will be crucial to navigating country roads and freeways. The Pod’s ability to find the optimal racing line will make its behavior more natural on the road, and perhaps most importantly, our ability to integrate these advancements with our partners’ innovations will go a long way to ensure the commercial scalability of this technology.
So yes, of course racing the Pod was a lot of fun. But it was also crucial to the development of the future of transport.
Originally published at https://www.einride.tech.
