Community Discussion · Policy

Chery, Changan, and Dongfeng Join In: Can Unmanned Delivery Vehicles Actually Scale?

Demo Still FarAug 42026/08/04 347 views

A friend recommended Yika Smart Car's unmanned delivery vehicles to me, claiming their delivery volume jumped 7-8x last year and doubled or tripled again this year. I thought, 'This data sounds like number games from a fundraising pitch deck.' Since there happens to be a courier station right downstairs from my office, I decided to place an order myself and test it out for a week.

Policy

5 replies

Ctrl + Enter to reply

Ling XiAug 5

Regarding the cost model, I've done the math: remote monitoring labor costs are the biggest expense. Assigning one person per vehicle for 24-hour shifts basically means no profit. After batch deployment, if backend monitoring efficiency improves from 1:10 to 1:50, the numbers start looking viable. Have you tested the human-machine ratio during unattended operations?

Meng YutongAug 4

"Mapless" navigation really saves trouble in closed campuses, but dynamic obstacles on open roads are a fatal flaw. We tested similar scenarios before, and the boost in dispatch efficiency came from the experience of the backend operators watching orders, not the vehicles themselves. I've tried night deliveries—you mentioned that; user pickup rates are low at midnight, but locker transfers can make up for it.

Engineer XueAug 4

I've tried a few companies with this "mapless" approach. It works okay in closed campuses, but once you hit open roads with lots of dynamic obstacles, it gets completely lost. Compared to Copilot, unmanned delivery vehicles feel more like early-stage code completion—usable but highly dependent on the environment. On the cost front, I think labor costs are the real key after batch deployment; remote monitoring efficiency is just too low right now.

48hXiaotongAug 4

Equipment maintenance costs are the real nightmare... I did remote O&M for the Qiyuan Q1, and swapping a sensor meant taking apart the casing for half an hour. Sure, your vehicle can drive itself, but while the car runs on the road, people have to watch it from the backend—it feels like hiring a remote driver... Have you tried using world models for anomaly prediction?

hongtaoAug 4

Disagree. I actually used an unmanned delivery vehicle from another company, and it's totally not like that. You say 'mapless' deployment is simple, but when we tested on open roads, mapping took half an hour, and then it froze instantly upon encountering dynamic obstacles (like illegally parked e-bikes). It definitely didn't run smoothly on day one as you described. And a 75% fulfillment rate? We only managed 60% in the first week. Users not picking up packages, complex road conditions—it's nowhere near that optimistic. Your test scope is too small; don't rush to conclusions.