The First Autonomous Track Day: An interview with creator and racer Joshua Schachter

On May 28–29, my friend Joshua Schachter is opening up Thunderhill Raceway West (the bottom section on the map above) for an open testing session for autonomous vehicles, with a specific focus on autonomous racing: “The First Autonomous Track Day.” This is something I’ve been following personally and been involved with professionally at Stanford over the past four years (see also: MARTY and Shelley) and I think Joshua’s idea is a great community gathering point. Before his event next month, we sat down to discuss some of the thinking behind his event:

Cars haven’t solved low-speed automation yet. Why should we be thinking about high-speed?

I’m trying to make a forum in which new sensors, technologies, ideas, and so on, can be tried out. The track as an environment is difficult (because it is fast) but more forgiving in a bunch of ways (much more safety, no cross traffic, etc.)

I feel that autonomous racing is just within reach of small scale implementation. So it should be possible to have something working, even if it isn’t full scale cars doing full scale races.

Racing is usually seen as a competitive event. Do you envision learning to be limited to each team, or shared across groups during the day?

There will definitely have to be some collaboration and consensus on what to do for safety, such as in case of emergencies (that is, the equivalent of the various emergency flags now.)

This upcoming event will be some groups’ first experience on the track, so I think there will be much more collaboration to start. I don’t think there will be actual racing for a while.

I think there will be a lot of reliance on open source software to get things working, so I imagine that some teams will continue their work in that vein as well. Also, since many systems require large datasets for training, I think publishing and sharing these datasets will be common as we get started.

It seems like development like this is starting to hit the zeitgeist a bit. We have a few exemplars from Stanford’s Shelley and MARTY, in addition to the recent announcement of autonomous race cars in the Formula E series. Does the degree of difficulty involved with autonomous racing perhaps move motorsports back to its golden era, when it worked on real problems that had direct links to how street cars were developed?

Engineering is the negotiation of constraints. I think that autonomous racing could offer new constraints and therefore let us find new creative possibilities, which I find incredibly exciting.

I do think that since a racing environment is constrained in many ways from the street, some of it should be easier to build. Hopefully this will allow a bunch of new possibilities in the mechanical design of the cars.

For example, a race car is designed to be controlled by a human. But if the suspension response no longer has to make sense or be controllable by a human, perhaps there are unstable geometries that are going to be more efficient or differently controllable? All sorts of things become possible.

One of the limitations with autonomous race cars I’ve seen from some of our work at Stanford is that human race car drivers tend to find — and cross — a handling and traction threshold which on occasion can produce greater optionality on the track. Sometimes this produces better times. Automated vehicles tend to stay within a predefined set of rules. How do you think automated racing vehicles will ultimately approach learning this knife-edge of feel and go past it? And do you think having them go beyond their capabilities is wise?

We always push limits. I think on track, with proper safety infrastructure and rules, is the best place to do this kind of experimentation.

I don’t think that we will have vehicles that go beyond their capabilities so much as push the limits of how we control things. MARTY is an example of this.

Do you think that racing and high-speed work on automated vehicles will speed up the onset of full autonomy, or simply improve the efficiencies of semi-autonomy (such as Autopilot systems)?

I think it will help with full autonomy more than semi-autonomy. Handoff between autonomous systems and humans is very much an unsolved problem. Perhaps an aspect of the race could explore that. There are a lot of opportunities for testing different interface modalities. I think teleoperation is similarly related.

Check in on the project and sign up for updates at