urbanMSP

A blog post about the Tempe Uber/pedestrian crash by an engineer who has worked on AV cars for Google.

http://ideas.4brad.com/it-certainly-looks-bad-uber

It’s a long read. Points: Working Lidar would have seen the pedestrian. Most other companies have two humans. The video shown is nothing like what humans would see or even good visual spectrum recording cameras. The Uber/Waymo lawsuit was over long distance lidar, could be tests around non-lidar systems. Modern cars often have advanced breaking systems that detect imminent crashes and engage to reduce damage but it didn’t happen as if it was disabled.

Feels good knowing pedestrians and bikers will be used as guinea pigs to work out the bugs in AVs.

Unrelated to the pedestrian discussion.

I had a thought this morning... it's pot hole season. As drivers many of us watch for the BIG ONE and will modify how we drive. Options are usually straddling a pot hole or slightly going over a line. Think we'll see an increase in blown out tires once the driverless cars don't look for pot holes?

Unrelated to the pedestrian discussion.

I had a thought this morning... it's pot hole season. As drivers many of us watch for the BIG ONE and will modify how we drive. Options are usually straddling a pot hole or slightly going over a line. Think we'll see an increase in blown out tires once the driverless cars don't look for pot holes?

Someone in NYC just told me their Uber car they rode in had some way of detecting and avoiding potholes - assuming its just something in a current car model

I heard Lidar has issues where light and dark shift, like edge of tunnel and its something they have been working on.

And please please, why was the car going 38 mph?!? who designs the car to do that - these are testing phase - they should be going as slow as possible without being hazard to other cars - like say automatically 5 miles under the speed limit on any surface street.

I’m thinking they haven’t solved pavement conditions yet or we’d see them testing these things outside of the desert. But in theory there’s nothing preventing a model from detecting potholes and avoiding them when safe. It’s not a fundamentally different problem from the other vision problems they’ve had to solve.

Unrelated to the pedestrian discussion.

I had a thought this morning... it's pot hole season. As drivers many of us watch for the BIG ONE and will modify how we drive. Options are usually straddling a pot hole or slightly going over a line. Think we'll see an increase in blown out tires once the driverless cars don't look for pot holes?

I thought about that as well as I crawled over potholes in St. Paul. If I'm using a car sharing service then frankly I don't care what happens to their car (unless I get charged for damages), but if I own it then I guess I'm not using it when potholes are prevalent.

Also, really stupid question, but lets say some stupid teenagers throw a bucket of water at an AV. Will sensors detect that its a liquid and drive through it, or will it slam on the brakes and give the rider(s) a heart attack? I can neither confirm nor deny that my friends did something similar when they were young, but with a bottle of sand.

The actions of the human "driver" in this case show starkly the dangers of Level 3 automation. The human driver was complacent with the technology and lulled into thinking her car was fully autonomous. It's an understandable, probably inevitable mistake, even if she was being paid not to make it.

Uber is going to have to answer some serious questions about this. It seems as though their LiDAR failed. Their human driver obviously failed. Their cars have caused one death in about 5 million miles driven, a rate about 20 times the fatality rate caused by human drivers.

What an awful company.

I’m thinking they haven’t solved pavement conditions yet or we’d see them testing these things outside of the desert. But in theory there’s nothing preventing a model from detecting potholes and avoiding them when safe. It’s not a fundamentally different problem from the other vision problems they’ve had to solve.

At the moment, seems more like "haven't solved yet."

I thought about that as well as I crawled over potholes in St. Paul. If I'm using a car sharing service then frankly I don't care what happens to their car (unless I get charged for damages), but if I own it then I guess I'm not using it when potholes are prevalent.

Also, really stupid question, but lets say some stupid teenagers throw a bucket of water at an AV. Will sensors detect that its a liquid and drive through it, or will it slam on the brakes and give the rider(s) a heart attack? I can neither confirm nor deny that my friends did something similar when they were young, but with a bottle of sand.

I mean, in general I feel like "resilient when water hits sensors" is a pretty core feature of AVs. Maybe the suddenness of a bucket of water could throw off the models that compensate for these things, but I would expect they'd solve a problem like that well before sending these things out without a human fallback. One thing to remember is that these things are dealing with loads of sensors of various kinds, and not just LIDAR alone, or cameras alone. The expectation is that in a situation where one sensor doesn't perform optimally, the others will pick it up. Just like different tools provide more or less useful input for certain factors (e.g., radar is great for figuring out how fast something is going, but useless for figuring out what that thing actually is), we would expect some subset of the sensors to work in moisture or partial obstruction, and for that subset to be sufficient for safe navigation.

As far as the dumb teenagers aspect goes...if they're smart, they'll learn from the experience with tech employee shuttles in San Francisco, and realize they need to plan for these things' high profile and controversial nature making them a magnet for vandalism and protesters. I'm less confident that they'll be sufficiently self-aware on this count.

I’m thinking they haven’t solved pavement conditions yet or we’d see them testing these things outside of the desert. But in theory there’s nothing preventing a model from detecting potholes and avoiding them when safe. It’s not a fundamentally different problem from the other vision problems they’ve had to solve.

At the moment, seems more like "haven't solved yet."

They have solved lots of technically complex and meaningful problems; the models they're able to create in real-time are already an incredible accomplishment in Computer Science. (Note that not all companies' AV prototypes are created equal, either.) It's just that there are still other problems to be solved before they can be commercialized. My point is simply that the modeling techniques they've used so far will probably adapt pretty well to a lot of the cases that haven't been tackled yet. There's a huge set of scenarios to be considered, and a much higher degree of certainty needed than we've ever demanded from machine learning applications.

https://www.bloomberg.com/news/articles ... y-concerns

Tesla downgraded by Moody’s. Bond offerings are now seven steps into junk rating.

Who could have foreseen their Model 3 production issues?

That was sarcasm, right?

Sent from my Nexus 5X using Tapatalk

A Tesla on Autopilot had a nasty one-car crash in Silicon Valley, killing its driver.

http://www.startribune.com/tesla-says-v ... 478457373/

Logs show the driver did not have his hands on the wheel. Autopilot isn't supposed to be fully autonomous and drivers are supposed to always be ready to take over. Between this and the Uber crash, I think we have a lot of evidence to support the position that humans don't make a particularly good fallback mechanism for AVs. I think it's human nature to lose vigilance when the car starts to feel sufficiently reliable.

The bigger takeaway is that truly autonomous vehicles are quite a bit further away than advocates like to say.

Sent from my Nexus 5X using Tapatalk

The bigger takeaway is that truly autonomous vehicles are quite a bit further away than advocates like to say.

Sent from my Nexus 5X using Tapatalk

I don't think these incidents really tell us that. I definitely am not confident in Uber's AVs. Tesla's is trickier because this was a normal driver in a production car; I would assume this doesn't represent the current state of their tech. But there are lots of others in the AV arms race with varying objectives, levels of expertise, and amounts of completed testing/training.

Personally I'm less concerned about the technological issues and more about the social and policy ones. There's going to be a huge range of quality and safety between the different providers of AVs and AFAIK we haven't set out minimum standards for safety and testing. We don't have a legal framework for assigning liability and fault. We haven't proactively set boundaries on how and where these things will share space with humans. We haven't really addressed the congestion problems they'll create. We don't have any regulations to protect user data privacy, or equity of access, or the potential consolidation of a meaningful chunk of the transportation industry (both manufacturing and service) into a few powerful players.

Personally I'm less concerned about the technological issues and more about the social and policy ones. There's going to be a huge range of quality and safety between the different providers of AVs and AFAIK we haven't set out minimum standards for safety and testing. We don't have a legal framework for assigning liability and fault. We haven't proactively set boundaries on how and where these things will share space with humans. We haven't really addressed the congestion problems they'll create. We don't have any regulations to protect user data privacy, or equity of access, or the potential consolidation of a meaningful chunk of the transportation industry (both manufacturing and service) into a few powerful players.

Great insight.

It’s like we’re at the beginning of the automobile age, and we have to make sure they don’t burn the streetcars and villainize pedestrians as “jaywalkers”.

Autonomous vehicle companies will pull all the same tricks as the car companies of yore.

The bigger takeaway is that truly autonomous vehicles are quite a bit further away than advocates like to say.

Sent from my Nexus 5X using Tapatalk

I don't think these incidents really tell us that. I definitely am not confident in Uber's AVs. Tesla's is trickier because this was a normal driver in a production car; I would assume this doesn't represent the current state of their tech. But there are lots of others in the AV arms race with varying objectives, levels of expertise, and amounts of completed testing/training.

Personally I'm less concerned about the technological issues and more about the social and policy ones. There's going to be a huge range of quality and safety between the different providers of AVs and AFAIK we haven't set out minimum standards for safety and testing. We don't have a legal framework for assigning liability and fault. We haven't proactively set boundaries on how and where these things will share space with humans. We haven't really addressed the congestion problems they'll create. We don't have any regulations to protect user data privacy, or equity of access, or the potential consolidation of a meaningful chunk of the transportation industry (both manufacturing and service) into a few powerful players.

All of that and they are already used as an excuse for not doing stuff.


Sent from my iPad using Tapatalk

PRT: No longer the car of the future.

Kind of sad, as I remember at the State Fair seeing the concept for it, but I'm not surprised. I don't know if this is an official designation, but the line in West Virginia is labeled GRT (Group Rapid Transit).

PRT may not necessarily be dead, but it would be very different from the mini monorail pods envisioned. I can see a similar system, but on the road with autonomous cars.

It does make me a little worried that buses, LRT, BRT, and/or commuter rail could have a similar fate but we'll just have to wait and see (and check how government from the local to federal level decides on transportation policies; whether to be more friendly towards transit, or the status quo of car-friendly, transit-hostile policies).

Oh it's not dead. Its legacy as the shiny thing anti-transit people use as an excuse not to do transit lives on in the form of autonomous vehicles.

Sent from my Nexus 5X using Tapatalk