Autonomous Vehicles
- Thread starter adam cool dude
- Start date
Interesting article about where Google is taking us in the predictive internet, including a recently filed patent. Author sees day when you are a passenger in an AV and you give a non-specific instruction and the vehicle performs based on your past preferences and its knowledge of you. The example in the article is "take me to a restaurant I would like".
https://www.computerworld.com/artic...tions-about-google-s-predictive-internet.html
https://www.computerworld.com/artic...tions-about-google-s-predictive-internet.html
FT (paywall protected) has an article today about the difficulties in creating and storing 3D maps for driverless cars. I found some of the points raised (summary below) of interest and helps explain why the introduction of these AVs could be slowed by these difficult issues that the industry is working on:
1. "One of the most important components for any robot driven car's journey is not just what it sees but what it knows beforehand about the area it is travelling through. The robot needs a map, but not just any map--these cars need a three-dimensional representation of the environment around them, one that is constantly updated and is accurate down to the centimeter."
2. "As it cruises through the streets, a self driving car collects more than a terabyte of data a day, enough to fill 1,400 CDs. With that much detailed info coming from the car's many sensors, hoever it is uneconomic to send it through a network like the internet. Instead companies have to physically move the data from one hard drive to another, a process sometimes called the "sneakernet" because engineers joke that the hard drives move at the pace of their own footwear."
3. "Researchers hope that eventually the base layer of info (data) will have applications not just for transport and logistics, but also for the development of augmented reality technologies---becoming like a simulation of the real world that can be used by any robot, drone or car."
4. "The reason these 3D maps are so important for self driving cars is not only for location but also because it reduces the amount of work that the autonomous software has to do to recognize the world around it. By comparing their actual surroundings to what was predicted in the map, the car can focus its attention only on things that are different, like identifying a pedestrian or a bicycle."
5. "Alphabet's mapping prowess is seen as a key advantage for its self driving car unit Waymo, which has already completed more than 4 million miles of autonomous driving. Alphabet owns Google Maps, Google Earth, Google Street View and the navigation app Waze, which tracks real time traffic."
6. "Within the industry there is a debate about whether to call these visual representations "maps" at all, such is their complexity. Accuracy is so important that even small changes, such as the shifting of tectonic plates that move a few inches a year, can have an impact."
7. "Regardless of what they are called, making these maps is extremely difficult. The huge volume of data used in the maps is one dilemma. Another is keeping them updated continuously so that they provide the latest info to the cars. Storage is difficult which is why autonomous vehicles are geofenced....they physically cannot fit the data in the trunk of the car, so are restricted to certain areas."
Companies are trying to deal with this problem by simplifying the map data so that it is easier to manage, but there is no single industry standard that has won out yet. Moreover, the AI needed to generate the maps is still far from perfect. Humans still need to check the labels on maps, assess the need for updates and analyze why the cars make mistakes during test drives.
Comments?
1. "One of the most important components for any robot driven car's journey is not just what it sees but what it knows beforehand about the area it is travelling through. The robot needs a map, but not just any map--these cars need a three-dimensional representation of the environment around them, one that is constantly updated and is accurate down to the centimeter."
2. "As it cruises through the streets, a self driving car collects more than a terabyte of data a day, enough to fill 1,400 CDs. With that much detailed info coming from the car's many sensors, hoever it is uneconomic to send it through a network like the internet. Instead companies have to physically move the data from one hard drive to another, a process sometimes called the "sneakernet" because engineers joke that the hard drives move at the pace of their own footwear."
3. "Researchers hope that eventually the base layer of info (data) will have applications not just for transport and logistics, but also for the development of augmented reality technologies---becoming like a simulation of the real world that can be used by any robot, drone or car."
4. "The reason these 3D maps are so important for self driving cars is not only for location but also because it reduces the amount of work that the autonomous software has to do to recognize the world around it. By comparing their actual surroundings to what was predicted in the map, the car can focus its attention only on things that are different, like identifying a pedestrian or a bicycle."
5. "Alphabet's mapping prowess is seen as a key advantage for its self driving car unit Waymo, which has already completed more than 4 million miles of autonomous driving. Alphabet owns Google Maps, Google Earth, Google Street View and the navigation app Waze, which tracks real time traffic."
6. "Within the industry there is a debate about whether to call these visual representations "maps" at all, such is their complexity. Accuracy is so important that even small changes, such as the shifting of tectonic plates that move a few inches a year, can have an impact."
7. "Regardless of what they are called, making these maps is extremely difficult. The huge volume of data used in the maps is one dilemma. Another is keeping them updated continuously so that they provide the latest info to the cars. Storage is difficult which is why autonomous vehicles are geofenced....they physically cannot fit the data in the trunk of the car, so are restricted to certain areas."
Companies are trying to deal with this problem by simplifying the map data so that it is easier to manage, but there is no single industry standard that has won out yet. Moreover, the AI needed to generate the maps is still far from perfect. Humans still need to check the labels on maps, assess the need for updates and analyze why the cars make mistakes during test drives.
Comments?
Those are really interesting points, it does show that outside of some niche situations or very small areas we are still quite some way off of having a fully autonomous fleet.
It's also interesting to look at things like penetration into the minds or the masses. I was out and about in MK the other day with my wife and we saw one of those little autonomous delivery pods mapping the pavements and she was fascinated by the thing like it was something from Star Trek. She found the concept of it being able to 'see' us and not cross the road in front of our car very strange.
It's also interesting to look at things like penetration into the minds or the masses. I was out and about in MK the other day with my wife and we saw one of those little autonomous delivery pods mapping the pavements and she was fascinated by the thing like it was something from Star Trek. She found the concept of it being able to 'see' us and not cross the road in front of our car very strange.
Good points. I think it is still early, pioneering days in the minds of the public as so little has actually been deployed in public spaces.Phoenix and Mountain View aside. But when AVs are deployed in well-mapped/geo-fenced areas those vehicles will look very normal. The first AVs to be commercially deployed by Waymo this year in Phoenix will be a ride sharing/taxi service that will compete with Uber and Lyft. By removing the driver, the ride should be priced competitively.
I was amused by a patent files by Waymo this week to cater to the queasy passenger. One of the concerns found in early testing has been that some passengers get car motion sickness. Check this patent application:
http://www.kyma.com/news/economy/waymo-patents-method-to-avoid-puking-in-selfdriving-cars/706529134
Big development in US: California has scrapped safety driver rules for driverless cars.
From 2 April, 2018, driverless cars go truly driverless as there will no longer be a requirement for a person in the vehicle to intervene in case of emergency.
50 companies, including Waymo, have a license to operate driverless vehicles in California. The new rules require the licensed company to operate the vehicle remotely---like a flying military drone--and to be able to communicate with law enforcement and other drivers when something goes wrong.
One of the main economic benefits praised by proponents of driverless vehicles is that they will not be limited by human boundaries and can do things like operate 24 hours in a row without a drop-off in alertness or attentiveness.
California joins Nevada in leading the move towards autonomous vehicles. Waymo has been testing AV without safety drivers in Arizona since October 2017. They plan to launch a commercial ride hailing service in Phoenix this year.
From 2 April, 2018, driverless cars go truly driverless as there will no longer be a requirement for a person in the vehicle to intervene in case of emergency.
50 companies, including Waymo, have a license to operate driverless vehicles in California. The new rules require the licensed company to operate the vehicle remotely---like a flying military drone--and to be able to communicate with law enforcement and other drivers when something goes wrong.
One of the main economic benefits praised by proponents of driverless vehicles is that they will not be limited by human boundaries and can do things like operate 24 hours in a row without a drop-off in alertness or attentiveness.
California joins Nevada in leading the move towards autonomous vehicles. Waymo has been testing AV without safety drivers in Arizona since October 2017. They plan to launch a commercial ride hailing service in Phoenix this year.
Why have they not banned people from driving cars then ?
There are hundreds if not thousands of fatal accidents with humans in control of vehicles, but no laws have been repealed or put in place to ban humans from driving.
Fully autonomous cars are on their way, far sooner than most will think. It is the future like it or not.
The UK will be soon be changing its laws to allow fully driver-less cars on public roads before 2021. Advanced testing of such vehicles is already going on, we have 8 cars now at work fully driverless, doing in excess of 3000 miles per week running 24/7 on our tracks and simulated urban environment, (minus refueling or recharging). All these cars could be ready for public roads with about a days worth of adjustments and reprogramming, so we are ready to go the moment the laws come in.
Everyone has to start somewhere.
The original horeseless carriage didn't start out with millions on the road immediately
But in the grand scheme of things it was not long before they took over.
Our tracks include over 47 miles of tarmac roads, dirt roads, pave, gravel roads, hill routes, handling circuits, high speed tracks, etc. The autonomous cars have full reign of any surface wherever they want to go, and are driving alongside plenty of other vehicles driven by our test drivers or other companies people.
We probably have nearly a hundred vehicles of one kind or another on track at any point in time, and we have not had a single incident or crash in the last 4 years.
One company we work for has also done extensive (over a million miles now) on all sorts of public roads all over the UK, not just motorways and a roads but country roads in England and up in the highlands and islands of Scotland, obviously currently they have a driver onboard in case of issues.
In fact they have 4 cars in remote Scotland this week doing extensive testing in the ice and snow, and by the reports I have been seeing the computers get stuck less often than the humans, as the computers have the patience to drive slowly and carefully rather than humans trying to go too fast too soon.
The big thing that people forget about autonomous cars is that, they all talk to one another, so what one learns in Scotland the test cars here on track then know an hour or two later.
We tested this out a few months ago at out test facility in Northern Finland. We had 4 cars doing various trials, 2 were in the lab and two out on track.
We set up a black ice scenario and the first car that came across it was going maybe a few mph to fast for the situation and did begin to slide very slightly, the second car went through about 30 mins later and also was a bit to fast and slid, both caught the slide and came to no harm or incident.
The two cars in the lab went out an hour late, and as they approached the black ice area they slowed and took more care, purely because they had uploaded information from the previous two cars saying there are traction issues in that area, so they went through the area without sliding or loosing grip in any way
Human drivers do not communicate between each other, so often you find many will hit the same black ice area and have the same issues, as they do not warn each other about issues they come across to make each others lives easier.
Entai's company AV testing looks impressive. I was particularly impressed with their ability to navigate black ice and communicate with the other cars in their fleet.
I am sure Entai will be impressed with Waymo's announcement today: 5 million miles and counting, on increasingly diverse road conditions. And their fleet too keeps learning from other AVs and simulated testing.
https://medium.com/waymo/waymo-reaches-5-million-self-driven-miles-61fba590fafe
I was very impressed with the video released today by Waymo showing a 360 degree AV riding experience and what it feels like to experience the Waymo AV technology. The drive feels totally ordinary for a passenger, even though it is technologically quite extraordinary. Clearly Waymo is on the cusp of launching self driving ride sharing with real people inside. This video is meant to reassure people and encourage them to participate.
Why is it so important to reassure the public of the safety of riding in an AV from Waymo? A Gallup poll published last week found that 54 percent of the 3,297 U.S. respondents said they were “unlikely” to use self-driving cars, with 59 percent saying they would feel uncomfortable riding in one. The survey was conducted in September and October.
After viewing the video, how likely would you be to ride in a Waymo AV in Phoenix Arizona?
https://www.youtube.com/watch?v=B8R148hFxPw
Why is it so important to reassure the public of the safety of riding in an AV from Waymo? A Gallup poll published last week found that 54 percent of the 3,297 U.S. respondents said they were “unlikely” to use self-driving cars, with 59 percent saying they would feel uncomfortable riding in one. The survey was conducted in September and October.
After viewing the video, how likely would you be to ride in a Waymo AV in Phoenix Arizona?
https://www.youtube.com/watch?v=B8R148hFxPw
Last edited:
Also these cars are all quite new. Computer tech tends to get less stable as the components age/degrade.
Would you feel safe in a car driven by a complicated 10+ year old computer? Who will test and replace computer parts and diagnose issues etc? Even now you sometimes have to replace ECUs on old cars and they are relatively simple and low power.
Would you feel safe in a car driven by a complicated 10+ year old computer? Who will test and replace computer parts and diagnose issues etc? Even now you sometimes have to replace ECUs on old cars and they are relatively simple and low power.
Last edited:
Main engine control units, controlling multiple functions simultaneously first appeared in BMW engines back in 1939.
Ok mainstream automotive ECU's as we know them did not really come about until the 1980's, but there are still lots of cars about with 20, 30 or even possibly some 40 year old computers in them all working perfectly fine.
If you have an issue you just replace the ECU job done no hassles.
Autonomous cars will be exactly the same, you just replace a computer and get the software sent down to it to get it back to exactly as it was prior to the box going down.
The main difference being in current cars you have one ecu controlling the engine, and or one computer controlling the abs, and or one computer controlling the suspension etc. One ecu or sensor goes down and you loose the entire system it controls.
in most autonomous cars I have worked on there are minimum two and often three or four backup sensors or computers on all the important systems.
These backup computers can be switched between as you are moving along if a sensor detects an issue with one it switches to another, you won't know anything about any issues, until you get a message from your garage to bring the car in for repair at some point., the car will still work and do it's job without any interaction from you as a passenger.
Ok mainstream automotive ECU's as we know them did not really come about until the 1980's, but there are still lots of cars about with 20, 30 or even possibly some 40 year old computers in them all working perfectly fine.
If you have an issue you just replace the ECU job done no hassles.
Autonomous cars will be exactly the same, you just replace a computer and get the software sent down to it to get it back to exactly as it was prior to the box going down.
The main difference being in current cars you have one ecu controlling the engine, and or one computer controlling the abs, and or one computer controlling the suspension etc. One ecu or sensor goes down and you loose the entire system it controls.
in most autonomous cars I have worked on there are minimum two and often three or four backup sensors or computers on all the important systems.
These backup computers can be switched between as you are moving along if a sensor detects an issue with one it switches to another, you won't know anything about any issues, until you get a message from your garage to bring the car in for repair at some point., the car will still work and do it's job without any interaction from you as a passenger.
It's not quite the same though, the AVs are full of sensors and they aren't cheap parts. They won't have backup sensors or wiring looms so there are still single points of failure.
Yea you can lose the ECU in current cars, but your there to still be able to stop and steer the car if something fails. Not just be a passenger as it goes haywire.
Yea you can lose the ECU in current cars, but your there to still be able to stop and steer the car if something fails. Not just be a passenger as it goes haywire.
As I said on the ones I have worked on, so I cannot say for all AV's, all have every sensor with at least one backup.
And as all the wiring looms on our AV's are multiplexed, then yes there is even a backup wiring loom.
If there is a signal error such as a broken wire, then the main computers can re-route sensor signals down another wire to avoid the broken wiring.
And you are correct, Yes sensors are expensive (currently) that is why the current cost of just one of the cars I have been working on is in excess of £2million each.
One the vehicles are fully productionised and economies of scale come in then those costs will reduce dramatically.
Also costs of sensors are dropping rapidly so that will get passed on in future.
There is not a single point of failure on anything electronic that has anything to do with safety, everything has at least one backup.
The default position of everything is stop safely, so if any sensor does go wrong and there is no backup or other safe method of carrying on the journey, the vehicle will just stop, carefully, so no one is at far less risk than in a normal non AV when something breaks.
Safety is the number one priority over and above everything.
I have read that the average cost of one of Waymo's AVs is currently $ 250,000, with dozens of these Fiat Chrysler Pacifica adapted SUVs currently on the road, esp in Phoenix, Arizona. Recently Waymo has ordered "thousands of Pacificas" from Fiat Chrysler to be delivered in 2018, so I expect economies of scale to begin to kick in. The economies of scale will help Waymo launch their commercial ride hailing service soon. It is assumed that these AVs will be in constant use, with the commercial service able to achieve profitability. Remember, AVs have no drivers, do not pay drivers (unlike Uber and Lyft), and since the cost of a driver is the single biggest journey cost to Uber and Lyft, it should be possible to make money.
As you state, safety is paramount and the video I posted yesterday again stressed this in order change the perception of the public to try out self driving cars from Waymo in the near future.
Did you have a comment on the video?
Think my comments are somewhat biased, as I am in and around AV's all day every day, so I am very used to them and the novelty has worn off somewhat from when I first sat in one, and was completely mesmerised by the steering wheel turning by itself and the car driving itself 3 years ago.
But then I am an engineer, so my overwhelming memory is wanting to know how it all works, and how can I make it better, rather than shock and worry about not being in control, the fact of not being in control has never been an issue to me.
I think it's really odd that everyone is talking about cars but no one is talking about lings like trains, trams and even buses.
We have the technology right now to fully automate these systems right now without even changing the infrastructure (signals etc). All existing trains could easily be retrofitted with all the tech and sensors needed to run.
I know the drivers unions will very much grind it out to the bitter end resisting this but there have been far too many avoidable fatal train crashes in recent years in the 'west' and they were all human error (driver or signaller). I can't really see an argument against it and the savings for someone like TFL would be astronomical.
We have the technology right now to fully automate these systems right now without even changing the infrastructure (signals etc). All existing trains could easily be retrofitted with all the tech and sensors needed to run.
I know the drivers unions will very much grind it out to the bitter end resisting this but there have been far too many avoidable fatal train crashes in recent years in the 'west' and they were all human error (driver or signaller). I can't really see an argument against it and the savings for someone like TFL would be astronomical.