Hacker News new | past | comments | ask | show | jobs | submit login
Apply HN: Zippy, self driving sidewalk meal delivery robots
22 points by tomjacobs on April 7, 2016 | hide | past | favorite | 27 comments
At Zippy, we're building self driving delivery robots that navigate the sidewalk to deliver meals from restaurants to homes. They use self driving technology including 3D cameras to go from a pizza/burrito place to the customer's front door, avoiding pedestrians along the way.

The robot cart currently is driven via 3G connection remotely over the internet, so it can be in another suburb over, with it streaming live video back. We've driven it down to the end of the street & back remotely. And we're working with GPS, sensors etc for SLAM. (We'd love to work with Varden Labs to cross-pollinate our self driving tech, that golf cart looks rad. http://abc7.com/technology/driverless-shuttle-giving-students-rides-around-mt-sac/1229716/)

Dominos recently launched a self driving pizza delivery robot in Australia, and other startups are building similar things. The time seems right for this new thing in the world for the first time. https://au.news.yahoo.com/technology/a/31135451/dru-the-new-self-driving-robot-launched-by-dominos/ https://www.starship.xyz

Home delivery via air drones sounds neat, but flying drones are dangerous. Navigating streets and foot traffic seems trickier than "as the crow flies", but the big plus here is that it's safe for the public, which makes it possible to do.

Having small items delivered to homes by robots gives us near zero cost per delivery, and deliveries are able to be done at any time rather than just peak hours; robots don't mind waiting around doing nothing for two hours unpaid.

            _____
  ___      |     |
    _-_-  _/\____|__\\__
 _-_-__  / ,-. -|-  ,-.`-.
    _-_- `( o )----( o )-'
           `-'      `-'
We're looking for skilled ROS and robotics engineers to help us make the future the present, help make startups like Sprig and SpoonRocket work out their unit economics, and help you get your delicious dinner at your front door with free delivery.

Video demos: http://meetzippy.com

Tom.




One problem with Zippy would be it contending with a myriad of laws stemming from whether Zippy is a vehicle, especially a motorized vehicle on a sidewalk. Let's ignore the unmanned part for now.

"Having small items delivered to homes by robots gives us near zero cost per delivery, "

You would still have variable costs like fuel, which would be incurred on a delivery by delivery basis. Then how high is Zippy's fixed costs?

I suspect that Zippy could only work in certain very densely populated areas, where users are willing to make concessions. In a suburb, Zippy because it using a sidewalk, would be too slow, but could do front door delivery. For an apartment building, Zippy could only go to the lobby's front door, hence requiring a concession from the customer that they may not want to make. Also how far can Zippy operate from the restaurant in question (it has to travel back)? What is its payload?

When I come to think about it, Zippy would be analogous to a person in a wheelchair making deliveries.

I suspect one could use computer simulations and some what-ifs on a spreadsheet to figure out Zippy's feasibility. I would wager that a person in car is simply more flexible, cost-effective and provides a more satisfying experience to the customer.


Yes, they're electric, so a depot with a solar array brings the cost down here.

Zippy hasn't worked out apartment buildings and the city, so is best suited at first to the suburbs. A few miles is that distance that seems silly to drive but too far to walk.

That's right, we already have manned wheelchairs roaming the streets, so it's not too different.

We bet there are cases where not having to wait on a human to be available will be a very good experience for the customer.


"That's right, we already have manned wheelchairs roaming the streets, so it's not too different."

However you probably don't want a person in a wheelchair delivering one's pizzas ....

Tom, again I would recommend you do some computer simulations and simple break-even cost analysis. Is the problem of pizza delivery really the human not being available, or being idle (is this really a capacity and demand management problem)? What stops a Zippy unit from not being available as it slowly heads back to base? From what I see, Zippy being a motorized land vehicle, suffers all the problems of a car, without any of its benefits. What makes drones appealing to say an Amazon, is the combination of flying that probably easier and safer than driving, and the value of the payload.

In general, last mile problems are tough. Also you have to ask yourself how does Zippy enhance the customer experience?

Cheers, Andrew


Right, it starts off somewhere with a clear closed path, and a daily need for small item transport.


My suggestion find a university to sponsor an on-campus student-to-student delivery network. Get the Engineering department or a robotics club to help with it. Find some sponsors like NASCAR does for the first one. other early adopters golf course communities.


I think something like this is going to happen fairly soon.

It seems like speed will be an issue if this is limited to sidewalks. Maybe what's needed is something partway between a car and say a bicycle. I think it needs to move faster than walking speed to be practical in most areas (you mention suburbs but delivery restaurants in suburbs aren't usually within walking distance of very many customers). I would think something more like the speed of a moped or bike would be needed. That means it probably would need to operate on the side of the road most of the time, not on the sidewalk. It would need to be quite narrow not to interfere with traffic. Maybe some sort of tricycle arrangement would be feasible.

As for actually building this you're going to need a team skilled in machine learning, computer vision and robotics and right now I don't think there are many of those around. You say you're looking for skilled engineers but what makes you think you'll be able to attract people with the right skills ?


I agree, it seems like this will happen in some closed area for a practical purpose soon, and then slowly expand out to more public areas.

A team with machine learning, computer vision and robotics experience is indeed what we need. People like this are currently in high demand, but there are more people quick to learn who are learning in these areas now.

If speed is an issue and it needs to travel by bike lines and road, perhaps some design based around this this or this will work:

https://www.youtube.com/watch?v=SPP7StCMYJg http://cdn.ndtv.com/tech/xiaomi_ninebot_mini.jpg http://www.motherearthnews.com/~/media/Images/MEN/Editorial/...


Ultimately: This. With a cooler.

https://www.youtube.com/watch?v=q2gehrAhflQ&nohtml5=False

https://www.youtube.com/watch?v=iyZE0psQsX0&nohtml5=False


I see two problems with this:

- not sure it is legal everywhere (in Europe for example bikes can't drive on sidewalk).

- robbers could just take the robot and keep what it was carrying.


Think of it like this: Don't steal people from the street; likewise, don't steal delivery robots who yell for help when you do.


Cool! I think this is a simple, good application of self driving tech. The self driving thing is neat but people (both restaurants and consumers) won't bother to use your tech for coolness factor more than a few times -- ultimately, everyone's going to adopt this only if it becomes cheap enough. So that needs to be your focus.


Yeah - cheap, reliable, and less problematic than humans. Then it crosses from novelty to practical daily.


So, how does this work in areas with shitty sidewalks? Or a lot of construction?

Also, how do you keep people from just mugging your delivery bot or trashing it?

Your videos are basically just cute little RC cars with sticks on them.

Also, what sort of equity are you offering your early engineers? Hopefully a lot, because it sounds like you really haven't thought this through.


Let's start with closed off areas and known paths. Then we can expand out.

Indeed they are, next up, self navigating RC cars, ala https://www.youtube.com/watch?v=pfJHgSNx_vY

Still looking for cofounders; so a lot.


What is keeping people on the streets from stealing food from your mobile delivery fleet? Or better, the tiny, cute robot car itself.


Don't steal people from the street.


Like you mentioned there are other startups doing this, and some of them are farther along eg. http://dispatch.ai/

what will Zippy do to differentiate from those robots? Why will Sprig and SpoonRocket buy a Zippy rather than a Dominos or Dispatch robot?


Dispatch definitely looks farther along but are they even targeting food delivery ? I just saw a techcrunch article about them that doesn't mention food but packages and mail.

Food delivery may require a somewhat higher speed, more road based than sidewalk based approach, especially if they are targeting cities and suburbs rather than campus-like environments.


Yeah, I don't see any of them targeting food specifically. Starship does groceries, which is close, and Dispatch as well.

I see more scooters on the road currently with meals in them than groceries.

DoorDash would never deliver you a single coffee from down the street. A robot would be happy to.

Bike lanes are the happy medium here.


There's enough market for everyone. Once it does the job of getting it there, with enough location coverage, other factors inform the decision.

Would you like your spicy calzone delivered in a grey box, or an Italian sports car?

Not saying it's all about the looks - but either way, robot logistics is coming.


This sounds really slow unless you can get it driving on streets. A reasonable maximum speed for a sidewalk is somewhere between 2 and 5mph.

What's to stop someone taking a delivery from the robot? Or from just taking the robot?

In what ways is this better than a human? How is this better than an airborne drone?


It's best for short distances.

The items are locked in a cooler bin (not in videos), and the robot acts as a human would if it were being kidnapped.

The robot is happy to run back and forth along a route all day, humans usually get paid for this. If you've flown a drone, you'll know that you don't want that anywhere near people in public. Think flying weed-whacker. Times 4.

What's really new here is that it makes it possible for the first time to have something physically be somewhere else in the neighborhood at the push of a button, with no employee or friend taking it there. Skype isn't teleportation, but we're getting closer.


Huh! How do i know the robot wont eat my meal?


This is one of our top problems we're working on.


Good. Solve that and i'll invest.


Hi Tom,

I've done a lot of work with autonomous self driving robots. Specifically relating to your company is the research I've done on an autonomous wheelchair that can navigate a sidewalk between two locations.

I know it's blatant self promotion, but I would encourage you to take a look at some of my lab's papers [1], which specifically focus on building the kind of system you describe.

One question I have for you is about your business model. Do you plan on selling these to a restaurant directly? Or do you plan on building a fleet of these things and licensing them out to restaurants (or basically any delivery service)? The answer to this question really informs your value proposition and the technology you need to develop.

For instance, you have emphasized that you've done some teleoperation tests, but this requires a human in-the-loop, whereas the value proposition for this devices is in removing the need for a human. If I have a human operating this thing, I might as well have him go out and deliver the food. So this points to the need for an autonomous system.

To really build a scalable business I think the best route would be to focus heavily on building the robot system. Do you have someone on your founding team who is a roboticist? You say you're looking for engineers, but I would say you need someone on the founding team who knows their stuff.

Here are some technical challenges that are still active research areas that you'll need to solve to make this an attractive option:

- Navigating without GPS. GPS is not accurate enough in cities or specifically on the sidewalk

- Large scale mapping. How do you make and store maps? You also have an opportunity here to use the delivery robots to constantly update the map and inform other robots of changes.

- Long-term autonomous operation. How do you recover from lost localization?

- Autonomous operation in different environments. e.g. indoors vs. outdoors

- Sensor calibration and characterization. You have an asus 3D camera on your demo car, but these fail in various lighting conditions and have limited range. You'll need better sensors.

- Crowd navigation. How do you predict the behavior of humans? How do you route a path through a crowd, a seemingly impenetrable obstacle, but one humans can navigate easily.

- Crowd localization. How do you figure out where you are in a crowd, when you have no reference to anything else?

And much much more. Anyway, best of luck! It's a problem I've put a lot of thought into, so feel free to drop me an email if you have any questions.

[1] http://vader.cse.lehigh.edu/publications.php


Thanks Corey,

Yes indeed GPS sucks on the sidewalk. Outdoor feature-based SLAM is the way to go. Congrats on getting the robot to go navigate 12K by itself with no loss of localisation.

The human operator is the backup, similar to Starship.

A separate mapping operation needs to be done first to pre-learn routes.

Recovering from lost localisation is tough and will need people to go recover the unit.

The Zed 3D camera operates better outdoors than the Asus, and solid state lidar sensors like Quanergy's are coming to market at a rapid pace and lowering costs. http://spectrum.ieee.org/cars-that-think/transportation/sens...

Crowds are tough too. The robot would not want to move too much in a crowd, and perhaps is best for it to just stay still until the crowd passes and it can re-localise.

These solutions and technologies are only going to get better with time, not worse, so it'll be a much easier problem given some short time.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: