UniTree Humanoid Companion
youtube.com
youtube.com
I think that the killer app for this kind of technology is a robot that can load and unload a dishwasher, and pick up dirty laundry, sort it, put it into the washer, transfer it to the dryer then fold it and put it away properly.
It doesn't need to be fast at this it just needs to be able to do this. A product that can meet that criteria and all the other things that this entails at the price point of a luxury car will sell like hotcakes to an aging Boomer population that wants to 'age in place.'
Where things start to get really interesting is where the robots have both the hand dexterity and a user serviceable design that allows one robot to fix another. From there a pair of robots with a sufficient amount of spare parts would in theory be able to build in the infrastructure necessary to build more spare parts and therefore could self replicate.
A self-compiling compiler.
Yes, but see also: https://news.ycombinator.com/item?id=44688151
To quote the relevant part of the tweet:
… Moravec's paradox within robotics: gymnastics that are difficult for humans are much easier for robots than "unsexy" tasks like cooking, cleaning, and assembling. …
Idea's been around for longer than people have have the nomenclature to describe DNA, let alone known that we had any or what it did: https://en.wikipedia.org/wiki/Self-replicating_machine
How long do you think it will be before a reliable dish washing / laundry doing domestic bot is on the market? a decade or two?
And how long after that do you think such a device will be capable of clanking self replication?
Depends if they only do this, or if they're fully general robots that also do this.
The former, remember we've already got dishwashers and washing machines.
The latter, to a standard that won't lead to regular news stories about how it tries to put a pet corgi into the bread bin or fry someone's pet goldfish for dinner, the problem space they need to navigate (literally and metaphorically) is much more complex than what is needed by level-5 FSD. Right now level-5 isn't a solved problem (running into parked cars should be "never happens" not merely "rare"), suggesting that even the more power-intensive compute (800 watts) isn't sufficient for the easier challenge of level-5 FSD, let alone domestic robots, and at that power level such domestic robots would need to be sold with an external computer running of mains not an internal one that runs of batteries.
Even if the software is a solved problem, which it isn't, getting the compute into a mobile body with the much smaller power envelope of an android vs. a car means the gap between level-5 and such robots is order of 5-10 years. But, to emphasise, this countdown has not yet started, and I'm not saying it will be between 2030-35.
For large-scale use, the global electrical power supply is currently 375 W/capita, and even with my expectation of wind+PV allowing us to double this by 2032, there's not enough electrical capacity for a huge number of these robots in this form on that timescale.
Regarding clanking self replication, that depends on the exact details of what you mean. It took a while to get the taxonomy of level-[12345] self driving, and right now we're too far from meaningful self replication to even make a taxonomy of self-replication:
• If all you need is industrial robot arms that work in a factory that makes more industrial robot arms, we've probably already got that.
• If you need the robots to be sufficiently general that we can fill up the payload bay of Starship with ~1000 of them, dump them all on the moon, and a few months later they've bashed enough rocks together to make an ore processing system, and a year later they've built a factory that makes more of themselves, that's much much harder and I don't think we've got enough information yet to even enumerate all the steps between here and there.