413 karma · joined March 24, 2016
IIRC, the lidar still lined up mostly because tree stems tend to not move, however, the larger problem was the error rate of the lidar sensor we were using. Readings further than 10m and the Hokuyo we were using tended to underestimate distances, so each scan of the forest looked a little but like the floor was curving over like that scene from Inception. Although maybe only 20 degrees. Still enough to be annoying.
Usually things that can kill people are triple redundant, where one goes bad, you can trust the other two readings for this reason.
I'm surprised there is a single point of failure like this in the design. But after I've read a bit about how Boeing released this product, maybe I'm not surprised.
For my thesis, I was trying to load two Street View photos side by side in a browser to compare people's perceptions side-by-side. Google maps at the time required you to load a javascript viewer for each image you requested. Think Hot-or-Not for cities.
The experience needed to instantly load a new image after the user voted because I knew they were only going to be on the site for maybe 30 seconds before they got bored and went back to reddit. I needed to collect as many votes as possible within that time period.
So I knowingly broke Google's ToS and prefetched the images on my server so I could provide the user experience I wanted. "I'm a small operation. Surely they won't know." I wrote a server side screen scraper to load the Street View images and exposed the scraped images with an API.
Now my site was faaaast. I could load Street View images instantly and in the end got over a million data points doing this.
But then one day soon after it stopped working. Then I got a cease and desist email from someone at google legal. They didn't respond to any requests for turning it back on or even to discuss. Radio silence. That was terrifying.
Since this was my thesis, I needed help getting my keys turned back on. Google in the end was very accommodating, but only after I used my nuclear option: asking lab director Joi Ito to bug Megan Smith while she was still at Google to help.
I was connected to some engineer and told them what I was doing and why. They said stop. But then a week or so later, they sent me a beta invite to their new Street View images api, where you can feed in a lat, lng, header to a query string and they'll just serve the image now. Pretty cool.
Offending site for the interested: http://pulse.media.mit.edu/
If you visualize technology progressing through time, there is an obvious merging of human and machine. Computers were large, now they are so small, you leave your Air Pods in your ears for hours because you forget they are still in.
Imagine another 50 years. Computers will most certainly be seamlessly connected to us in some way.
Imagine a device that can read your mind and give you what you want before asking? What if you're more intelligent just because you buy a specific product? What about a device that is proven to lengthen your life just by using it?
Biological research and health are what's necessary to get there.
The interesting things happen at the intersections. Who could compete?
Steve seemed like a smart guy. Maybe if he saw things this way he would bet the company as you say.
We originally approached the customer service problem from this angle, but found our approach better suited to our vision.
They are great and I've used them before.
And for further clarification, your bill would probably be slightly higher than the minimum, but quite honestly, we wouldn't feel comfortable taking you on as a client for quite some time.
As we grow and move downstream, we'll lower prices and expand our offering. It sounds like you're just downstream; nothing to get upset about. Just wait and eventually we'll make sense for you too.
On the back end, I wish I could brag a little about this, but it is our "secret sauce" so I'm just going to have to say it's proprietary and I believe the beginning of something great.
To directly answer your first question regarding edge cases, it's simple: we don't do them, opting instead to let your in-house team do the complex edge cases. Over time though, we inevitably end up doing some of them as we find commonalities across clients (shipping exceptions, insurance disputes, etc).
Now might be a good time to clarify our pricing a bit and state that our pricing is not a one size fits all model. We analyze our clients ticket and call histories and find a price that works for both.
The $999 is a minimum spend and from that we deduct cases and voice minutes. The minimum covers training and overhead to ensure we're not losing money for clients that are too small. We've also found that, in our experience, companies who tend to spend less than $999/month are sometimes still at the point where their support tickets are full of a lot of edge cases and are best handled in-house.
We're aimed at small and medium companies, not individuals running small projects (yet). Someday in the future hopefully, but we're not there yet.
That said, keep in mind there will always be a lot of repetitive time consuming stuff, and that's where we excel. Everyone likes to think they're a unique and irreplaceable snowflake, and that may be true in total, but there are parts that can be handled more efficiently elsewhere.