A shared tech stack doesn't mean a shared set of risks

Momenta put Robovan on the road in Suzhou. The official line is it's built on the R7 world model, shares a tech stack with Robotaxi, and meets automotive-grade requirements. These sentences read smoothly. My first reaction is: sharing a tech stack saves money, true, and the risk is real too.

In quant this is called out-of-sample. You make a beautiful backtest curve in one domain, then switch to another domain, the feature distribution changes, and decay is common. Robotaxi's domain is urban passenger transport, Robovan's domain is logistics freight. The routes are more fixed, but load, loading/unloading, nighttime hours, and stops are all different. Automotive-grade solves hardware and safety processes. Distribution shift, it doesn't solve.

I wrote about the Stelato V8 a while back, which was also about the distance between pretty parameters and real scenarios. Same here. The official line says Robovan is meant to expand commercial scenarios for scaled unmanned business—I believe half of that. Scaling depends on pushing down per-kilometer cost. And in per-kilometer cost, the labor share of safety redundancy and remote takeover—logistics and passenger transport are two completely different ledgers.

One more small thing. The name Robovan is also used by Tesla, and the trademark involves an earlier applicant company. Doesn't affect the tech, but when doing competitor tracking it's easy to mix up data. When building tables you need to keep a column to distinguish them.

So right now I don't recommend extrapolating Robovan's commercial value using Robotaxi's logic. The Suzhou deployment is worth watching, and Momenta and Uber's plan to test in Germany next year is worth watching too, but that's data from another domain. To really judge, wait until it runs out its own operating metrics—daily volume, takeover rate, per-kilometer cost—not the four words "shared tech stack."