I have thousands of dollars to spend on something like this if it had an arm and could do things like get items from the fridge by verbal order (with no step by step programming, just visual recognition of the fridge and how to open it), play catch with the dog, sweep up, act as a security guard when I'm gone, etc.
Robotics is one of those things where the hardware, price, and networking are there but the software isn't. We don't have an AI-lite engine we can toss in for simple things a dog could understand like "get my slippers." Until someone cracks that code, this stuff is just going to be rich-boy novelties and unneeded contenders in the already over-saturated telepresence market.
Batteries aren't good enough, linear actuators have awful power/weight ratios, and computers just aren't fast enough to solve CV problems and calculate grasps and paths in seconds, rather than minutes.
Saying you "have thousands of dollars to spend" is great, but not sufficient. The PR2 I talk about in the blog post costs four hundred thousand US dollars, and it sucks! It's like saying you're willing to spend five thousand bucks to buy a Lamborghini Aventador. It's going to be many decades before a useful household robot only costs ten grand. A household robot just requires too many breakthroughs in too many different fields.
Oh I don't know, it seems to me that the PR2 is designed for industry so its pricing is going to reflect that. I suspect there is a home robot space that some startup can fill sooner than later. Whether it does all the things I listed is the big question and I suspect it won't, but it may be able to do a few things that make it a worthwhile purchase.
I've played a bit with opencv, pcl, ros, etc. There's some very impressive image recognition stuff available right now that works well on commodity x86 platforms(1). I don't think the market is expecting a HAL-like or Jetsons like robot, but I could see something akin to an early 80s home computer where the product is clearly a long list of compromises but it does a few things very well and is compelling. Home robotics may be the same way for a while until it has its 1984 Macintosh moment, which as you say might be a decade or two or four away.
I did appreciate your posting, but I think its a little dismissive of the some of the homebrew and smaller scale stuff out there. The PR2 is a VC backed monster designed to bring industrial robots to retail, hospitals, etc. These guys want to build the 747 of the robot world. That's great. But there are people out there building the Cessnas of the robotics world. I expect an affordable consumber product that isn't a joke xmas 2018-2020. There's just way too much potential here.
1. Home robot hackers falling in love with the super low power NUC which gives a CPUMark score around 5,000+
The PR2 has an arm payload of 1.8kg, total payload of 20kg, and a top speed of 1m/s. That doesn't sound "industrial" to me, that sounds like "the absolute bare minimum to be a mobile anthropomorphic robot". And to achieve those numbers, it weighs 480 kilos! That's a payload fraction of 4.1%! How is this anything like a 747?
If it only weighed a 60 kilos, the average weight of a human, then we would expect a total payload of 2.4kg, and an arm payload of... 7.3 grams. Doesn't sound too useful to me. And the damn thing would still cost $50,000!
The PR2 was a VC backed monster, but remember, Willow Garage went out of business last year, because their products just weren't very useful! The technology just isn't there, and won't be for a long time.
Home robot hackers falling in love with
the super low power NUC
I'm sure that's fine for pathing, and Kinect-SLAM, but "getting a beer from the fridge" is picking arbitrary items in arbitrary poses in an unconstrained environment, basically the Amazon Picking Challenge, which nobody can solve with reasonable speed yet, even with hundreds of thousands of dollars of equipment.
If you honestly think you can build a cheap robot that can do all that by 2020, then by all means, launch a startup and earn billions of dollars. But I don't think it's going to be done before 2040.
Robotics is one of those things where the hardware, price, and networking are there but the software isn't. We don't have an AI-lite engine we can toss in for simple things a dog could understand like "get my slippers." Until someone cracks that code, this stuff is just going to be rich-boy novelties and unneeded contenders in the already over-saturated telepresence market.