For the past few years we’ve heard a lot about self-driving vehicles. Experimentation and testing have been going on for years and will likely continue for a few more years. Although self-driving cars aren’t quite there yet that doesn’t prevent people from making hyperbolic promises. One thing that strikes me is that it’s surprising that we’re not seeing self-driving race cars being developed.
Why (Not) Self-Driving Race Cars?
If AIs can surpass humans at chess, game shows, Go, and other things then why shouldn’t we expect them to beat us at driving? Remember how the world watched in fascination to see whether an AI could beat the world chess champion? Or how well it would do playing Jeopardy? That would be nothing compared to a watching a robot race car taking on the best racing drivers in the world.
Safety issues are the obvious problem preventing “robot race cars”. Car makers are still working on self-driving cars that operate at road speed, let alone at racing speeds. However, this may not be as big of a challenge as it sounds. Driving in a race is a significantly different problem to solve than driving in a race. Races are simpler, more controlled environments than the normal road environment. For example, you don’t have people with strollers walking out onto the track whereas that’s a real problem for road driving. Furthermore, a real challenge is being able to react quickly, and appropriately, during the change conditions of a race. This is a potential strength for AI. But people will have little tolerance for a robot race car making a mistake that hurts or kills someone, so there’s little room for error.
The Future is Here: Roborace
An organization exists, Roborace, that is in fact running races of AI-driven race cars. These races are closed to the public for safety reasons so are covered via livestream. They have a 12-race season, which they call “Season Beta”, where they race several vehicles. The vehicles are called “DevBot 2.0” but the AI driver software varies by team – each team trains their own racing models. The vehicles are electric, running on the NVIDIA DRIVE platform when in autonomous mode. They can also be driven by humans, enabling the teams to explore assisted driving capabilities. In short, Roborace sounds like a great R&D environment for car companies as well as a lot of fun for everyone involved. But there’s room to do a lot more.
How Will Self-Driving Race Cars Come About?
Here’s how I see this playing out:
- Manufacturers test with limited races of their own cars. We’re almost seeing this with the Roborace folks right now. I have no doubt that some of the car manufacturers are doing this within their own R&D labs too.
- A manufacturer runs a public race with their own cars. More realistically, they’ll hold a race and record it, then release the recording as long as there isn’t a spectacular crash during the race. I suspect we’ll see this before the end of 2025.
- One car manufacturer challenges the others to a race. I’m looking at you, Elon Musk. Any such challenge could come at any time, but the race itself is unlikely to occur until the 2025-2027 time frame.
- AI will be used to augment human race drivers. AI is already being used to develop F1 race cars, to analyse and identify the strategies that lead to better driving performance. AI is also being used to help train F1 drivers via simulations. As the Roborace people are doing, I have no doubt that F1 drivers will soon take advantage of AI augmentation if they’re not doing so already.
- Self-driven race cars will compete against humans. Self-driving race cars competing against human divers in Nascar? In F1? This is definitely a few years away due to the significant liability challenges. My guess is that by 2025 Nascar will pass a rule against self-driving race cars given their audience. F1 however, being more R&D focused, will likely be more circumspect and not pass such a rule. If that’s the case, then my guess is that by 2030 a self-driving race car will win an F1 race.
All of these dates are my own gut feelings based on exactly zero insider information, so take them for what they’re worth.
The Potential of All-Robot Racing
Just like people love watching “Robot Wars” they may like to watch Robot Races. The entertainment value would be incredible. Consider several scenarios:
- Races by individual cars. This would be an all-robot race, run similar to Nascar or F1 rules. I’m not convinced that this will hold much interest for people except in the early days when it’s a novelty.
- Aggressive “races” by individual cars. The idea is that the race cars are allowed to bump and even take out other cars in order to win. This would basically be “robot wars racing” and would likely prove to very entertaining if done right.
- Races by teams of cars. A variation on strategy #1 and is similar to team bicycle racing where several racers work together as a team in order to win.
- Aggressive races by teams of race cars. With this strategy it’s teams of robot race cars against other teams of robot race cars, a variation on #2. Limits would be set, in particular the maximum number of cars in a team. The teams would be allowed to choose their own strategy. Perhaps there’s one “race car”, two defenders, and one offensive car in a team. Done right the entertainment value would be incredible.
Is AI the Future of Car Racing?
Building an AI-driven car to compete solely against other such cars is a much simpler, and safer problem, to solve than an AI-driven car built to race against human-driven cars. The former problem speaks to the advantages of AI, speed and reaction time, although the latter speaks to our base human desire to compete. I believe that self-driving race cars are inevitable in some for. Time will tell.