Comment by traverseda
5 days ago
So the play here is obvious, use the teleoperation as training data for a more general purpose AI controller. You need that data to make a model in the first place.
What doesn't make sense to me is the cost. Yes, $8000 is probably low for this robot but it's a reasonable price range for something like this. The AI credits though? I know vision LLMs are not cheap, they're not going to run something like Llama3.2vision on every frame. Very curious about the embodied AI architecture that this is going to use, and how it can get cheap enough that it's not going to use $500/month in electricity every month.
The world of computer vision is much bigger than multimodal LLMs. You'd run an ensemble of specialized models for 3d mapping, object classification, path validation, and so on. On a raspberry pi 5 8gb you can run what you need to self drive an RC car on an obstacle track at 10 FPS.
Today, a lot of it is just integrated VLAs. End to end backprop covers a multitude of sins - a single integrated model stack in cross-attention beats "an ensemble of specialized models".
The cost is really tough to pin down, yeah.
The way one of their employees told me it to me as like a dishwasher.
Of course the dishwasher should be more expensive. When you add up the hours in labor saved and multiply by the hourly median wage, you get something in the $50k to $100k range.
But it's essentially just a sprinkler.
Ain't no one going to to pay the cost of a new BMW for a dishwasher.
Same thing here for the laundro-bots. Their competition isn't against the time saved for a person to do it themselves. The competition is a maid that does your whole house for $70.
> Of course the dishwasher should be more expensive. When you add up the hours in labor saved and multiply by the hourly median wage, you get something in the $50k to $100k range.
Except most dishwashers aren't competing against people washing dishes by hand and making the hourly median wage, they're competing against other dishwashers
It may actually be cheaper to have the robot remotely operated all the time from some low cost country...
>> Very curious about the embodied AI architecture that this is going to use, and how it can get cheap enough that it's not going to use $500/month in electricity every month.
In practice most of the costs would be the teleoperation costs rather than AI inference.
8k is cheap if laundry is fully offloaded but will a regular consumer spend 8k on a device that is not proven? I guess there is a subset of consumers that this automatically targets/caters to.
At what point is it actually cheaper? Laundry isn't that expensive to do yourself, or to outsource if you really don't want to do it yourself.
I'd pay $8k tomorrow for a bot that would 100% do my laundry. That means collecting it from the various dirty clothes hampers throughout the house, bringing it to the washer and dryer, operating the washer/dryer, folding and putting clothes on hangers, and putting them back into the dresser and hung up in closets.
For a bot that just automates an in-house laundry service that washes and folds? Not very interesting since it might save maybe 60% of the time, but practically zero percent of the mental overhead.
This seems like a step towards that I suppose. My house isn't configured to make it an option even if it was a fully-baked product, but if these ever get to the point of actually working without remote teleoperation I'd certainly be in the market.
17 replies →
You’re not replacing the outsourcing it component though, you’re replacing a maid at home doing it for you. In home laundry services are a very different experience since you don’t have to also go pick up and drop off the laundry.
A service like that can be hundreds a month, so pay off period is on the order of years… which could be worth it.
1 reply →
I doubt it will ever be cheaper than doing it yourself, like most things in life. The market is for people unable or unwilling to do it themselves.
Outsourcing can be difficult and expensive in many regions. The lack of an actual human might even be considered a benefit in some cases, such as nursing homes (although you have to weigh the benefits of human contact with the benefits of fewer humans spreading plagues).
Set price too low to be truly rare luxury show off item, but high enough that expendable income is necessary for first movers. Trade in kind by "gifting" to influencer types: the pop science tech nerd ones to legitimize it by scrutinizing current downsides, the effortlessly luxurious ones to establish it as a brand, and a few mom-core ones to seed the aspiration). Develop better versions from the initial data, drop prices a few times a year via holiday sale or via model deprecation, keep current model pricing high. Develop 3rd Gen and introduce "pro" tier. Very tried and true strategy (many step omissions of course) and imo they nailed the price point for initial show off. It's not really affordable for its market but it's also not unaffordable if you consider the costs of what it would replace if it turns out to work!
Tesla operates vehicles for $100/month. I’m guessing whatever cloud ai this thing needs is less complicated and less money.
Tesla runs its stuff on ~150 watts of local compute that's bundled into the price of the car. The $99/month is just to rent the software.
> So the play here is obvious, use the teleoperation as training data for a more general purpose AI controller.
Strong disagree: the play here is to use teleoperation, claim it's AI, make a shit-ton of money and cash out before the house of cards come tumbling down.