Hacker News Reader

Comment by sandov | Hacker News Reader

>It’s going to be some time before we have the machine intelligence necessary to do enough for this robot to find a toolbox on its own, retrieve the necessary tools, unload the drywall from a truck, carry it to the room in question, install it, and complain about OSHA all autonomously.

I don't think it's necessary to develop machine intelligence to do some of those tasks. The toolbox could could emit EM pulses so the robot can find it, the tools could be put in a specific order and put back the same way by the robot, the drywall could have easy-to-remove arrow stickers so the robot knows how to find it's borders, etc.

View on HN

Two and a half years ago I wrote a paper on "robotic automation" for a big financial services company. One of the conclusions was that, while the world of things was heretofore designed with humans in mind, in "the near future" (i.e. about where we are now), "stuff" would be designed for "machine-first" use, rather than "human-first" use.

This isn't surprising. We've seen this before in the transition from "animal engines" to "steam engines" (think - the whole infrastructure around the care and feeding of the animals etc transitioning to manufacture, fuelling and support for steam engines). We're starting to see it now in the (slow) transition from "carbon-fuel transport" to "electric transport".

> One of the conclusions was that, while the world of things was heretofore designed with humans in mind, in "the near future" (i.e. about where we are now), "stuff" would be designed for "machine-first" use, rather than "human-first" use.

This.

Just to provide an example, humanity already has self-driving cars for decades, and they are deployed and are extensively used in public transoortation. The trick to solve this problem was to not force human-designed solutions to an automation problem.

Thus, instead of trying to automate vehicles to run on roads, we have vehicles running on railway tracks.

Problem solved.

That actually brings up an interesting concept that never occurred to me.

We could get automation of complex tasks far earlier than I had thought. We're at this divide where we're getting capable with our understanding of motion and mechanics, but intelligently figuring things out is a massive roadblock.

What if, as you say, we make very dumb robots where their interaction with the world has been entirely laid out by humans. Ie, a humanoid robot making toast seems difficult. Yet if we give it the exact dimensions of the room, rf sensors for locations of shelving, cabinets, bread, etc, it seems possible with our current tech that the robot could be programmed to handle dynamic environments as long as the inputs were laid out ahead of time.

There's limited value in something like toast making, defining the inputs would be a hassle. Yet, for house building, or factory automation, if every piece, shape, location and dimension was known it seems a fleet of robots could build a house with quickness and ease, no "intelligence" needed.

Humans would need to lay a foundation to a spec of dimensions, but then the bots could go nuts. Neat!

What you're describing sounds like the way pre-fab homes are built. A computer model is created which specifies every part, and that gets sent to the factory floor where all of the wood is picked, cut, and assembled into pre-fabbed shipable components. As much work is done in the factory as possible while keeping all of the parts flat so they're easy to stack onto a truck. The cutting is all done by CNC machines, the assembly is all done using jigs and much of it could be done by assembly-line-style robots, moving stuff around can be done by simple robots as well.

The components get stacked on a truck by humans running machines, but the stacking order is pre-determined by the computer model, so this can potentially be automated by an oversized version of common manufacturing machines. The truck is driven to the job site by a human, but that might one day be done with an autonomous vehicle.

The final step is assembly on-site, which is done by humans running big machines. This is potentially the only part where humans need to be heavily involved, to make sure everything goes correctly together and to do the fine adjustments and small details.

The translated source document implies that they already mark the drywall with QR codes. Considering all the other markings I've seen on drywall, this should be trival to add at the factory.

Reply on news.ycombinator.com