It depends on how fine-grained we're talking. If you rely on your map for absolute perception (every object must exist in your map in its current real-world state), then yes, you're in trouble. If you rely on the map as a basic prior, and perceive the world through different means, then a map (to me) is the right way to go. You avoid a whole category of computer vision in doing so.
The good news is that a point cloud map is relatively trivial to create, and the pipelines to update such a map over time are becoming more understood. There's lots of startups taking on this challenge.
Correct! We think people would be surprised at how big some of these private communities are, and how many places look like them other than retirement communities.
We are an optimistic bunch, and it will be a slow rollout (constrained by vehicle build-outs), but we also have experience launching such self-driving taxi services.
We have a service that's live today in San Jose that serves tons of (repeat) riders daily. We've learned a lot there that we're applying directly to Florida.
How autonomous is the service in San Jose then? Do you mean to say that you have cars without drivers that serve ride after ride and come home to charge? If so I'm far behind the curve.
We still have Test Drivers behind the wheel as a last resort, but we regularly give passenger rides with zero or low disengagements.
As a real-world example: we have a couple who uses our self-driving taxi service multiple times a day every day (The Grims), and their rides are ~95+% autonomous.
So how do you get rid of the human backup driver? Until they go, this technology has negative cost-effectiveness. Remote manual control at slow speed for difficult situations? Dispatching a backup driver?
Hello everyone! Oliver, the CEO of Voyage and YC alum (2011) here.
I got started in the self-driving car world as a member of the team that built Udacity's self-driving car curriculum, and also an open source self-driving car. This industry is crazy fun and moving ridiculously fast.
Let me know if I can answer any questions at all about absolutely anything.
We're really excited about bringing Level-4 transportation to the world, but especially so at places like The Villages, Florida. Whether it’s helping those with Parkinson’s, Alzheimer’s, vision impairment, or just those who want to get around with less friction, we’ve seen first-hand the positive impact autonomous transportation can have for seniors.
1) Yes, at least today. We call them Operations Specialists, and they're highly trained in operating autonomous vehicles and customer service.
2) This is a tough one, since it's not apples-to-apples. That said, Waymo is clearly the industry leader in this regard, and we're glad that at least some of their formula is observable from the outside. We've been at this a little less than a year, and today many of our passenger rides are 95+% autonomous. A long way to go, though!
3) We love these communities for a number of reasons:
A) Speed limits in these communities tend to be a little more restricted than everyday public road. The faster you (and others) can go, the more complex it becomes. The roads are also spectacularly well-maintained.
B) Customer acquisition is a non-trivial problem, and with these sorts of partnerships we gain (exclusive) access to 125,000 potential riders. It would be brutal for us to achieve the same in San Francisco, and not necessarily cost effective to retain them.
C) If we wanted to try out concepts that may accelerate the timeframe to remove the Test Driver, a private community is easier to collaborate with.
D) Above all else, there's just a lot of demand for what we offer in these sorts of communities. Self-driving cars will lower the cost of transportation, and for those with vision impairments (or Parkinson’s, Alzheimer’s etc.), it enables such residents to get around with less friction.
> 2) today many of our passenger rides are 95+% autonomous
Do you mind if I ask what portion of your rides are ~100% autonomous? I feel like that's a more interesting metric for progress toward removing human operators.
>> (1) Will these cars have a Voyage employee in them to supervise?
>>>> 1) Yes, at least today. We call them Operations Specialists, and they're highly trained in operating autonomous vehicles and customer service.
My understanding of the autonomy levels (for what they're worth) is that while a human is behind the wheel, autonomy is not at level 4. If a human is required to take control in certain situations, it's either level 3 or 2.
I have a feeling human operators are there for more customer service purposes. Given the location selected it might be wise to have staff on hand. The idea of no humans makes me think of the news story recently on the front page here about the hitchhiking robot being destroyed also seems relevant, although private areas may be less prone to incident like this.
(https://www.wired.co.uk/article/hitchbot-usa-vandalised-phil...)
Also according to this techrepublic article, while level 3 may still require a driver, a level 4 vehicle can still be limited to the "operational design domain". So human drivers can still be in place for extreme edge cases. Although I don't know the specifics of these vehicles, and don't think the "levels" of driving are very useful.
Thanks for the info. One more question if you have the time: to what degree are you limited by data, vs. other factors? If you could have 10x the training data, would that 10x your performance, or are other factors in the way (eg. compute, neural net architecture, etc.)?
We design our algorithms to not be necessarily reliant on more data for better performance. We recognize that fleet size and large-scale data collection just isn't going to be a key strength for Voyage, so we design our algorithms accordingly. For example, our classification stack required only 9,000 point cloud frames to achieve pretty amazing performance.
1) It's not an immediate priority for us, and not really a strength. We are partnering with an amazing company for our second generation vehicle that will go a long way towards this, though. In general, we are excited to see LIDAR sensors become slimmer.
2) One of our engineers builds and maintains Carloop.io, which I can verify as awesome!
3) Safety and performance. One project to keep an eye on is ROS2, which is prioritizing development of real-time features (RTOS, DDS support).
> We are partnering with an amazing company for our second generation vehicle that will go a long way towards this, though. In general, we are excited to see LIDAR sensors become slimmer.
I grew up near The Villages. Really crazy to see such a cool innovation basically coming to my home town. That being said, I don't know if self-driving cars are the right approach in that market. Have you considered self-driving golf carts? =P
(for those of you who don't know, The Villages has arguably more miles of golf cart paths than highways, and all shopping plazas have golf-cart sized parking spots, so most of the retirees cruise around on their carts)
Hi Oliver, this is a really cool announcement. I have some questions:
1) What are some of the issues surrounding adoption in local municipalities that you've faced with the deployment of this, specifically in Florida?
2) How do you go about convincing a local municipality to change their laws or issue waivers regarding legal issues about vehicles on the roadway that are autonomous, as specifically in Florida, the legal framework does not exist yet that I'm aware.
3) Explain your pricing model for the taxi's in The Villages and also does pricing decrease once the vehicles become fully autonomous (no driver in the front seat required)
I'd personally like to see a day, in my lifetime, where the pricing model of autonomous transport rivals that of vehicle ownership (which in my math is about $0.20 - $0.50/mile)
re: "Whether it’s helping those with Parkinson’s, Alzheimer’s, vision impairment, or just those who want to get around with less friction..."
I can see the positives of less friction for this segment of the population. Kudos.
That said, what effort has Voyage put into helping the local community (and law enforcement) prepare for this?
For example, a person with Alzheimer's gets lost...now what? True, this might happen now but if you're successful then the social infrastructure might not be ready, willing and able to handle the add demands on that system.
I'm going to hang up and take my answer off the air :)
We take our responsibility with any passenger _very_ seriously. We strive to be world-class in how we operate 100% of the time, and we learn everyday from the best customers we could find as to how we should handle situations like these.
No one has figured this out yet, so we have some ways to go, but rest assured we take our responsibility super seriously.
At Voyage we're working to bring about the end-goal of self-driving cars: a world where anyone, anywhere can summon a car directly to their doorstep, travel safely to their destination, all for an extremely low price.
Crucial to our mission is motion planning. To travel from A to B requires a series of decision making and motion planning in the chaotic world of urban driving, and we need world-class engineers to work on this key problem.
This code is legit (see other comment; download the tar and try to forget you ever visited Sourceforce) and incredibly interesting. I browsed this repo a year ago and still come back to it every so often. The architecture decisions that the Stanford team made in 2005 and 2007 have stood the test of time (at least as evidenced by many of the decisions in the architecture of our Voyage car, but anecdotally in others too).
Hi Oliver, I have enrolled for the Udacity self-driving car Engineer Nanodegree and cannot wait to start! Do you think that browsing through this codebase will help me better understand some of the concepts taught on the course?
The good news is that a point cloud map is relatively trivial to create, and the pipelines to update such a map over time are becoming more understood. There's lots of startups taking on this challenge.