From https://angel.co/kiwicampus: "Current self driving algorithms require huge amounts of data to be trained with, so our robots are initially being controlled from Colombia."
(not affiliated, I just found these DIY-like robots very fun)
No! :)
*we do continue to label data though and will probably have a large "AI Trainer" staff in place for any NEW skill we want Amy to have in the future.
" so our robots are initially being controlled from Colombia." There was a sci-fi movie about that[0]. In short a wall gets built between the US and Mexico, and telerobots controlled by workers in Mexico start being used for the jobs normally performed by migrant workers.
That being said there are some interesting legal issues with telerobots. Currently, most laws we have presume the person performing the crime is physically present. For example, in the unlikely event that one of these robots injures someone or causes property damage who exactly is held liable?
Another interesting case is if an evil set of Berkeley engineering students in desperate need of robot parts for a project, pays the driver of one of these robots to disable tracking and drive it to a secret robot chop-shop. Supposedly, the penalties for the evil Berkely engineering students can be less than if they went out and kidnapped the robot, because the company allowed and trusted the worker with access to the robot.
I live in Berkeley in one of the areas that they're testing these and there's 20+ restaurants that do take-out within a mile of me, so that part isn't a big issue. Even a mile could be a 30 minute round trip though, as they run on sidewalks and so inherently can't really go much faster than walking speed, and that's not very many deliveries per hour per robot.
Autonomous robots are far cheaper than humans to operate and could more easily parallelize the task of doing multiple deliveries per hour. In addition, the markets will expand significantly when the price of food delivery comes down to $1-2. Also, car-driven food deliveries also only average about 2 per hour, in part because of parking issues on both restaurant and customer end.
(disclaimer: I am a co-founder of one of their competitors -- robby.io -- we're based in Palo Alto/Stanford)
I agree - it's more like saying something like this may happen in the future so we'll put out a low-end prototype and you let your imagination do the rest. For robotic food delivery to be competitive a lot of things are still missing.
Other than the video camera that's presumably recording everything around them and the real-time location reporting? If you just want to damage it you can chuck a brick at it from its blind spot, but that's true of everything.
I wouldn't be surprised if communists and Antifa would destroy them on purpose in the name of fighting for "working class" and jobs. And if not them, then other thieves/vandals do the job.
Wow - that site is terrible. On initial load I have 3 ads (left side, right side and bottom) that take up 10% of screen each, cover the content and other ads, and scroll with me.
I recently switched to JavaScript by whitelist only and now I won't go back. I always thought disabling JavaScript was just for tinfoil-hat luddites, but now I realize how often JavaScript detracts from the experience and how little it enhances it. Speed-wise, it reminds me of going from 56k to DSL back in the day.
I hate to be so negative but this looks terrible. I remember seeing that other food delivery within San Francisco that uses a robot but at least it looks like there was real engineering put into it.
We are actually using monocular computer vision to navigate in the sidewalks and in some robots LIDARS for collision avoidance, we are also testing with SLAM detection in the smartphone we have in our kiwibots. We have made a great progress in automation and self driving, also in network software for take control of the robot if we need it (last 3 minutes of every delivery we control the car to make sure we deliver in the correct address, also for crossing streets and crowded sidewalks. Still in a semiautonomous phase but really excited about the future, any feedback to improve our product?
I almost tripped over one of these last week! Then I stared at it as it rolled away, wondering if they were being remote controlled or had some basic AI and pathfinding. Sounds like a little of both.
I think it's a great start, but the clearly have a lot of work to do. They should start by adding a pole and flag so you can see it coming. :)
Completely unrelated, but thank you for posting the chicken papricash recipe on your website. Sounds really good. Just printed it out, am planning to try it.
You're welcome. :) Not sure I wrote it on there, but you can substitute plain yogurt for the sour cream for a smoother taste. Also, my brother sometimes adds a little powered sugar right before blending for extra sweetness.
At a different delivery robot company, some of their positioning hardware doesn't work all that well in practice, and each robot is followed (on foot) by a 'robot minder' that corrects its course using a videogame controller if necessary.
Regardless, it's probably cheaper to outsource driving via remote control than to employ domestic delivery drivers... or building robust self-driving vehicles
Redwood City has Starship Technologies robots doing Doordash deliveries in the downtown area. Those look more like a finished product. They're mostly autonomous but can be controlled over a cell phone link if necessary.
There is something unsettling about the prospect of fleets of
bomb-defusing robots wheeling around pedestrian sidewalks on their unknown delivery missions.
If these become ubiquitous, they'd also provide a ready cloak for anyone wishing to detonate a bomb without any link to the crime or chance of harming themselves.
This reminds me of hearing Reed Hastings of netflix talking about starting the DVD mail-rental business as preparation for when streaming became technologically viable. They knew that internet delivery would eventually work, but to be the big first mover they needed to establish a customer base. As soon as broadband reached a level where streaming video is viable, they convert the customer base over to it.
Likewise, robot street delivery is going to be huge. The marginal cost of delivery will tend towards zero, and its really useful. If a company spends years and zillions of dollars messing around with AI and machine learning bullshit to power these, they will miss the boat. The goal of a successful company should be:
- Get as much VC as possible
- Spend it on low cost chinese built bareboned GSM controlled bots
- Hire warehouses full of vietnamese computer gamers to control them
- Get more VC
Once you have a market cap of billions, THEN you get a lab of engineers together to make them increasingly autonomous
So compete at a high burnrate in an extremely crowded space with a huge workforce faking your technology until you can pivot your product and market down the road?
I get what you're saying as a general heuristic; but this is not that in my opinion.
Here it's important to remember that the only reason that worked for Netflix was because they could be a huge disruptor even before streaming. They IPO'd, and basically all but put Blockbuster out of business, years before they launched streaming.
Netflix could've never moved to streaming and still made a lot of people a lot of money and have been a successful business.
But if they'd been losing tons of money on DVDs, that 10 year wait between launch and streaming would've been much, much more interesting.
Hi, Im the cofounder of kiwicampus.com, thanks a lot for the feedback, we are actually using monocular computer vision to navigate in the sidewalks and in some robots LIDARS for collision avoidance, we are also testing with SLAM detection in the smartphone we have in our kiwibots. We have made a great progress in automation and self driving, also in network software for take control of the robot if we need it (last 3 minutes of every delivery we control the car to make sure we deliver in the correct address, also for crossing streets and crowded sidewalks. Still in a semiautonomous phase but really excited about the future, let me know if you are interested in know more about us.
40 comments
[ 2.5 ms ] story [ 73.2 ms ] threadFrom https://angel.co/kiwicampus: "Current self driving algorithms require huge amounts of data to be trained with, so our robots are initially being controlled from Colombia."
(not affiliated, I just found these DIY-like robots very fun)
This was the story I was referring to: https://news.ycombinator.com/item?id=11520681.
Was it false, or did things change in the past year?
That being said there are some interesting legal issues with telerobots. Currently, most laws we have presume the person performing the crime is physically present. For example, in the unlikely event that one of these robots injures someone or causes property damage who exactly is held liable?
Another interesting case is if an evil set of Berkeley engineering students in desperate need of robot parts for a project, pays the driver of one of these robots to disable tracking and drive it to a secret robot chop-shop. Supposedly, the penalties for the evil Berkely engineering students can be less than if they went out and kidnapped the robot, because the company allowed and trusted the worker with access to the robot.
[0]https://en.wikipedia.org/wiki/Sleep_Dealer
I don't see food from outside a mile or 2 km range being deliverable in any reasonable amount of time.
Is this the food-delivery equivalent of the last mile (kilometer) problem? Because with delivery the rest of the miles are also important...
(disclaimer: I am a co-founder of one of their competitors -- robby.io -- we're based in Palo Alto/Stanford)
http://imgur.com/a/TKJNC
https://techcrunch.com/2017/04/12/marble-and-yelp-eat24-star...
This is just a glorified rc car with a basket on top being controlled by outsourced workers. Not even a MVP prototype imo.
I think it's a great start, but the clearly have a lot of work to do. They should start by adding a pole and flag so you can see it coming. :)
Regardless, it's probably cheaper to outsource driving via remote control than to employ domestic delivery drivers... or building robust self-driving vehicles
[1] https://www.youtube.com/watch?v=7y3-AmqVg5E
If these become ubiquitous, they'd also provide a ready cloak for anyone wishing to detonate a bomb without any link to the crime or chance of harming themselves.
Likewise, robot street delivery is going to be huge. The marginal cost of delivery will tend towards zero, and its really useful. If a company spends years and zillions of dollars messing around with AI and machine learning bullshit to power these, they will miss the boat. The goal of a successful company should be:
- Get as much VC as possible
- Spend it on low cost chinese built bareboned GSM controlled bots
- Hire warehouses full of vietnamese computer gamers to control them
- Get more VC
Once you have a market cap of billions, THEN you get a lab of engineers together to make them increasingly autonomous
I get what you're saying as a general heuristic; but this is not that in my opinion.
Netflix could've never moved to streaming and still made a lot of people a lot of money and have been a successful business.
But if they'd been losing tons of money on DVDs, that 10 year wait between launch and streaming would've been much, much more interesting.