Technology

I bought a $300 fake Roomba thing. It was on clearance.

And i fought against it for years. But ended it up coming in clutch for a lot of reasons.

It did not have an app, just a IR controller. Its pretty dumb. It bumps into everything. It gets stuck under things. I sometimes have to create a maze so it cleans a specific spot.

Its been a habit of mines to avoid anything with an app that requires internet access. But the product lines are shrinking, and I know at some point, if I want a Roomba, I'll need to invite always-on AI or whatever.

It's funny how these "smart" appliances are all addressing things radically important for households, but in a poisoned way from the beginning. As if those making them were just trying to get there first and win the bank.

There's a problem of scale in industrial innovations, where bigger scale makes cost of production of something and cost for the consumer and network effect power better, meaning that there's no market feedback to help those who came first get old and die to make space for those who come next.

I think this tendency is actually the solution - there is a feedback, it's that lacking feedbacks on one level prohibits those undying monsters from being competitive on the next one. The niche of non-poisoned smart appliances won't be filled by anything big, for example.

That's also another funny moment - instead of dedicated appliances it makes it useful to have one universal one (basically a butler robot) that can be programmed. It's an incentive in the direction of universal machines programmed by customers.

BTW, imagine a frame with various manipulators and sensors attached to an RPi via GPIO, where every manipulator/sensor can be whatever thing at all, just needs to have a manipulator/sensor description template. The OS of the RPi itself runs tasks of the "move those items of fragility categories such and such to such and such locations, remove dust and dirt from that surface, wash that window", for which the existing set of manipulators/sensors and task sequence are optimized without user's involvement (other than attaching them and providing the right description templates, though I suppose manipulator controllers can provide them too, and confirming the resulting jobs). That's also where those LLMs etc are good enough, to interpret instructions and display the sequence of actions they are going to perform to get user's confirmation. This way you won't have to fear that you tell it something harmless and it starts a fire in the room.

Such a system needs a set of standard protocols for the sensors\manipulators, their description templates, and the representation of commands deciphered from human speech to a set of tasks, and the spaces and traits of objects. The programs visualizing the resulting offered set of tasks, deciphering the order, optimizing one set of tasks into a better one, and so on, should be pluggable. Suppose everything's already made, just nobody really needs a thing that they can't just buy and turn on.

OK, I like imagining, should work better instead and start my toy the weekends after the next ones (I suspect I won't start it even by then, at least not in the initial ideologically good form ; nothing about robotics or home appliances). Spent these weekends on making a POV-Ray scene instead.

Why did I even write this.