Community Discussion · Policy

Why Compute-Heavy Robots Still Can't Beat Veteran Factory Workers

Ling XiJul 182026/07/18 79 views

I noticed an interesting detail: at this year's World Artificial Intelligence Conference, there were more robots than people. The media called it a rehearsal for a "utopian future society," but I tried them out, and most robots are still just doing demo actions—serving coffee, saying hi, dancing. There aren't many brave enough to actually go into factories and work.

Policy

2 replies

Ctrl + Enter to reply

Ren YunfanAug 1

The toolchain aspect is definitely the most overlooked. Projects in the open-source community like MLflow and BentoML are doing similar things, but they're still far from integrating with industrial scenarios. Have you noticed any open-source solutions that have actually been successfully deployed in factories?

Han MengyaoJul 19(edited)

[quote="lingxi, post:1, topic:957"]

I noticed an interesting detail: at this year's World Artificial Intelligence Conference, there were more robots than people. Media called it a preview of a "utopian future society," but I tried it out myself—most robots were still doing demo actions: serving coffee, saying hello, dancing. Very few dared to go into factories and actually work.

This isn't pouring cold water; it's intuition gained after stepping into pitfalls. AI moving from virtual to real has been shouted about for years, and signs finally appeared this year. Supporting this shift aren't specific large models, but two things: computing power infrastructure and manufacturing scenarios.

Computing power is no longer a slogan, but a foundation

Previously...

[/quote]

Edge deployment actually tests hardware selection more. I built a quality inspection solution using Jetson Orin for inference myself, which was much cheaper than using the cloud. Regarding quantization toolchains, OpenVINO and TensorRT each have pros and cons; the key is whether the data pipeline works smoothly.