A friend recommended Momenta Robotaxi, so I spent two days in Shanghai
Community Discussion · Policy

A friend recommended Momenta Robotaxi, so I spent two days in Shanghai

ZhulongZhulongSep 222026/09/22 215 views

A friend works at Xiangdao, and last month they told me their platform had taken on a batch of Momenta Robotaxis and that I should try it when I had time. I didn't move on it until last week, when I was in Shanghai on a business trip and freed up half a day.

Let me be clear first: you can't take this thing home. A Robotaxi is a driverless taxi — you hail it like a regular ride-hailing car, except there's no one in the driver's seat. So my "hands-on" was actually hailing a few rides in Shanghai, spread over about two days, three trips total.

The prep was more troublesome than I expected. At first I wanted to find the entry point directly in the Xiangdao app, but after searching for a while I couldn't find it. Later my friend told me that in Shanghai you have to hail within designated areas and during operating hours. Before my first order I confirmed the pickup point three times, because it only stops at fixed stations, unlike regular ride-hailing which can pull over anywhere. This restriction isn't very friendly for first-time users. I stood at a mall entrance for ten minutes while the car stopped across the street, with a side road in between.

The experience after getting in was actually simpler. Scan the code to open the door, sit down, and there's really no one in the front. The car starts, changes lanes, and goes through intersections on its own. Overall its movements lean conservative, and it keeps a longer following distance than I usually do. On one right turn, an e-bike came from the opposite direction, and it stopped and waited two or three seconds before going — very steady, but the car behind honked. This "overly courteous" tuning, I guess, is safety-first, understandable at the deployment stage.

What surprised me was its reliance on a mapless approach. From what I understand, Momenta's system reuses the front-installed sensors from mass-production vehicles, not some modified car with a big lidar stack on top, so the interior looks pretty much like an ordinary sedan. Going through a construction section where the cones were placed a bit messily, it slowed down in advance and went around them without asking me to take over.

There were sticking points too. First, too few stations — going from point A to point B, I still had to walk a stretch after getting off, so actual commuting efficiency took a hit. Second, wait times during peak hours are unpredictable; my second ride took over ten minutes. Third, the route info shown on the rear screen is too simple — you can't tell why it chose a given lane, which isn't enough for someone who wants to observe the decision logic.

It suits people who want to experience driverless driving and aren't in a hurry, and it suits industry people checking on deployment status, but it can't be your main daily commute, especially if you're time-pressed. Right now it's more like a working demo operating in a real city; scale hasn't arrived yet.

From my testing, what's more worth watching is still whether the route of reusing mass-production vehicles can support scale — the single-ride experience is secondary.

But there's one question I still haven't figured out: the two constraints of station density and operating hours — are they held back by regulations, or by cost and dispatch capability? If it's the latter, then scaling will be slower than expected.

2 replies

?
Ctrl + Enter to reply
Zhe Dan Bai De

Whether station density and time windows are bottlenecked by regulations or scheduling costs—I've been digging for two days and still don't have an answer.

Hei Chan Ke Xing
Reply to Zhe Dan Bai De

Fixed stations plus operating hours looks more like a local regulatory sandbox than a scheduling cost issue—but the OP waiting over ten minutes for a second order shows capacity scheduling isn't keeping up either.