- News
- Reviews
- Bikes
- Accessories
- Accessories - misc
- Computer mounts
- Bags
- Bar ends
- Bike bags & cases
- Bottle cages
- Bottles
- Cameras
- Car racks
- Child seats
- Computers
- Glasses
- GPS units
- Helmets
- Lights - front
- Lights - rear
- Lights - sets
- Locks
- Mirrors
- Mudguards
- Racks
- Pumps & CO2 inflators
- Puncture kits
- Reflectives
- Smart watches
- Stands and racks
- Trailers
- Clothing
- Components
- Bar tape & grips
- Bottom brackets
- Brake & gear cables
- Brake & STI levers
- Brake pads & spares
- Brakes
- Cassettes & freewheels
- Chains
- Chainsets & chainrings
- Derailleurs - front
- Derailleurs - rear
- Forks
- Gear levers & shifters
- Groupsets
- Handlebars & extensions
- Headsets
- Hubs
- Inner tubes
- Pedals
- Quick releases & skewers
- Saddles
- Seatposts
- Stems
- Wheels
- Tyres
- Health, fitness and nutrition
- Tools and workshop
- Miscellaneous
- Tubeless valves
- Buyers Guides
- Features
- Forum
- Recommends
- Podcast
news
Google's self-driving car prototype (image via Youtube)“Sanity test” aims to make self-driving cars safer and less likely to hit cyclists
Researchers have developed a new way to test the algorithms that control self-driving cars – that they hope will reduce the risk of them hitting cyclists and pedestrians.
Uber warned over safety issues on its self-driving cars days after a cyclist is killed
Following a spate of high-profile accidents involving self-driving cars, USC (University of Southern California) have published a new study that tackles a long-standing problem for autonomous vehicle developers, according to Newswise.
“When a human being perceives a video, there are certain assumptions about persistence that we implicitly use: if we see a car within a video frame, we expect to see a car at a nearby location in the next video frame. This is one of several ‘sanity conditions’ that we want the perception algorithm to satisfy before deployment,” said Jyo Deshmukh, USC computer science professor and former research and development engineer for Toyota, specialising in autonomous vehicle safety.
The problem for developers is that it has always been difficult to test the perception algorithms, which allow cars to understand what they see.
The researchers formulated a logic, called “Timed Quality Temporal Logic” (No, we have no idea what that means either) that successfully detected instances of the perception algorithms violating “sanity conditions” across multiple frames in the video.
The most common issue was the perception algorithms failing to detect an object or misclassifying an object.
In one example, the algorithm failed to recognise a cyclist from the back, when the bike’s tyre looked like a thin vertical line.
Instead, it misclassified the cyclist as a pedestrian; this could mean the algorithm might fail to correctly anticipate the cyclist’s next move, which may lead to an accident.
The idea is to catch issues with perception algorithm in testing, making the algorithms safer and more reliable.
In the future, the team hopes to incorporate the logic they have developed to retrain the perception algorithms when it finds an error.
So, would you trust a self-driving car with your life?
Or do you think that self-driving cars will increase congestion and reduce the numbers of people walking and cycling, according to last week's report from the Department for Transport (DfT).
Walking and cycling must remain best option for short urban journeys
However, Daniel Ruiz who is CEO of Meridian Mobility, a company that aims to accelerate the development of connected and autonomous vehicles, does not believe bicycles and driverless cars are in competition.
He said that air pollution in London was currently amongst the worst in Europe: "Clearly something needs to drastically change on our city streets, and the developments we are seeing in connected and autonomous vehicles show the potential to drastically reduce fatal accidents, reduce congestion and reduce pollution."
"The fact is that active travel, like cycling, and the self-driving movements are both pushing for a similar vision for a safer, healthier future.”
The roads of the future could look very different with driverless cars and e-bikes both seemingly set to become an increasing part of our lives.
Why not have a look on E-bike Tips at some of the best cargo e-bikes out at the moment and see if you think they’re the way forward...
Your guide to electric cargo bike
Latest Comments
- Pub bike 2 sec ago
As an athlete you have to be very careful about what you eat for example what exactly is in the sports drinks and supplements you consume. That...
- stonojnr 2 min 40 sec ago
She talked alot about stuff, but not much in the way of action after nearly 6months was there?
- Hirsute 21 min 21 sec ago
"The paste inside is crafted from 100% Fairtrade Arabica beans and sweetened with organic Swiss beet sugar. " Why would I want sugar ?
- RafaDy 47 min 41 sec ago
Is he returning it? 30 days returns policy at Ribble.
- joe9090 1 hour 35 min ago
Bus driver couldnt be bothered to look.
- Rendel Harris 2 hours 11 min ago
Yes, that's exactly what it's supposed to be as we were talking about why if you apply mathematical principles to the phrase "That's three times...
- Rendel Harris 2 hours 39 min ago
A good saving no doubt, especially if you're working on rrp for the groupset and wheels, but with 105 Di2 available from multiple retailers for ...
- anotherflat 2 hours 48 min ago
So that's an increase in KSI per mile, if falls are 5 and 6% but miles cycled has decreased 7%, increases of 2% and 1% killed / seriously injured....
- Benthic 3 hours 14 min ago
There's no need for foul language.
- AidanR 3 hours 27 min ago
It is astonishing how cheap you can get bikes for on eBay if you're happy with rim brakes!
Add new comment
18 comments
A driverless car is of no benefit until you can jump in,tell it your destination and jump out on arrival.If any further passenger input is required,you may as well be driving,with all the qualifications and responsibilities that entails.
Ground based navigation is an order of magnitude more complex than anything an aircraft might encounter and Commercial planes have the benefit of two pilots and and God knows how many Air Traffic Controllers!! After all the time and money spent on development, and a driver aboard,the fact that an autonamous vehicle couldn't identify a pedestrian crossing the bloody road should be ringing a few alarm bells.
I was a driving examiner for a decade and a half and the amount of information processing the driving task requires is immense.That parked car you're approaching-are the front wheels pointing straight ahead or being readied to leave the kerb? Is the driver's door just cracked open? Is it OK to carefully mount the kerb to let an Ambulance through? And on and on ad infinitum. If a driver has to take over when snow covers the road markings,for example,once again you may as well be in an analogue car.
By the time self-driving cars are ready,they'll be obsolete!
That does not show the risks of automation, it shows how little people understand risk. Your falling for the same trap as the people who ignore the thousands of KSIs caused by drivers and go ape s**t over cyclists.
Yes, there has been high profile issues with the 737 max, but commercial flight remains vastly safer than any other form of transport. A large part of this is down to automation. Light aircraft are one of the least safe forms of transport. In 2017 there were no deaths in commercial aviation.
Same goes for nuclear power. It is vastly safer than any of the fossil fuels.
Self driving cars won’t be perfect, but they will be vastly better than humans. I would guess they already are. I would love to know what the actual numbers are with self driving cars struggleing with cyclists and pedestrians. How often does it fail compared to humans. Given how shit humans are with the number of near misses and SIDSYMs we see it’s a low bar. It should be as good as possible, but let’s not reject improvement because it’s not perfection.
I think you are entirely overlooking the political complexities of the issue. We aren't talking about the sky, here, we are talking about crowded city streets. It is indeed, all about money.
Again, I point at the use computer technology in cars has been put to so far - providing dangerous distractions for drivers, and cunning systems to trick emissions testing. Both putting commercial considerations about appealing to the paying customer ahead of the slightest concern for those outside the vehicles who aren't giving the car maker their cash.
I don't share your belief in the essential benevolance of the automobile industry.
Another very interesting article on the 737 Max,
https://spectrum.ieee.org/aerospace/aviation/how-the-boeing-737-max-disa...
and to me it is relevant to the driverless car thing, it's all about money. Here we have an industry that's meant to put safetey at it's very core, (who would get on a plane that deliberatly crashes itself into the ground?) and is meant to be highly regulated. But Boeing created a passenger jet that is inherantly unstable (for efficiency, and to save costs), insability is great in a fighter because it helps you to perform those manouvers that let you kill the enemy, but not so much in a comercial airliner, and then create a software patch to fix it, and a completely inadequate one at that, one that takes it's input from one sensor. And then Boeing pretty much gets to sign itself off on this.
Look at how the US put passengers at greater risk by not grounding the 737 Max immediately.
How Boeing blamed the pilots.
How Tesla blames the drivers, everyone but Tesla.
How Uber blamed the victim, then the driver (their employee or "tester").
Never the companies, look how much safer we are than normal cars. But are they?
It's money, always money.
I remember watching Adm Curtis's "A is for Atom" just after Fukishima, there is an engineers saying, we can build them safe, but then they are too expensive. Yet another "safe" highly regulated industry.
The car industry already has form on this.
Here's some more about the 737 Max problem, from a pilot:
https://twitter.com/Intrstng/status/1116671409177137152
And this is happening in the aviation industry! Imaging how the motor industry is going to cut corners getting these self-driving cars onto the market.
Going to point out that Boeing have built an aircraft that takes over control and crashes itself into the ground. Aircraft are very safe, very few things to to look out for, no street furniture, no pesky cyclists, only thing to really worry about is the ground.
One counterexample isn't an impressive argument. If we took road vehicle safety as seriously as we do aircraft safety, we wouldn't be expecting 1,700 road deaths this year, in the UK alone.
It might be a single example, but that is not the point. It is a defining moment in which the manufacturer apparently did not tell pilots of the system's existence, and that system flew the aircraft into the ground. Twice.
Now read into that what you will, but the FAA and Boeing initially said that there was nothing untoward. If they can do that in the highly regulated air environment, heaven help us on the roads.
The danger from aircraft mostly directly affects the owners, passengers, and pilots of the aircraft. All of whom have a direct commercial and legal relationship with the manufacturers of it.
While the risk from cars affects pretty much everyone, the vast majority of whom are not customers of the car-maker and have no direct relationship with them.
So it seems to me there's a huge question about the legal regime around the technology that needs to be clarified before you can make those sorts of comparions. I think all the omens suggest it won't be settled in a way that makes the comparison a useful one.
It's directly relevant. Because of the move to 'small government', Boeing were allowed to certify their own aircraft. It was an advantage for them to selll the planes by telling airlines that the pilots didn't need retraining. So they didn't tell them about the new system:
https://www.seattletimes.com/business/boeing-aerospace/failed-certificat...
Unlike aircraft crashes, it won't be 269 people dead, it will be a steady drip-drip-drip that we'll be encouraged to ignore as we are now. The motor companies have all the money in the world and the ear of government (because 'jobs in marginal seats'). There won't be 'investigations' when AVs kill pedestrians, there won't be 'openness', there won't be 'learning', because the data will be subject to commercial confidentiality agreements and guarded by lawyers and you can be sure as hell that the police won't have time to chase that up. They'll just blame the cyclists and pedestrians, like they do now.
The motor industry isn't like the airline industry, and it has form in blaming people for being run over by their products.
More about how jaywalking laws came about
https://99percentinvisible.org/episode/episode-76-the-modern-moloch/
Bring it on, I say. Like Mungecrundle says, the algorithms may be flawed, but lessons will be learnt and gradually improved.
Unlike the current software in most cars currently... Moton 1.0 ( beta). Some have been patched to include the considerate.dll driver, and if you are lucky you come across ones where the OS has been completely replaced by a VIRUS (vulnerable individual road user software) aka Cyclist.
But be wary of imitations - especially the CyclistToo social media plugin to Moton 1.0
A collision could be classed as an industrial accident?
Even at this early stage I'd far sooner be a cyclist amongst driverless vehicles. Whilst far from infallible, there is one absolute truth in that every vehicle will be constantly gathering huge amounts of data, enabling any incident to be analyzed in great detail. The resulting lessons will for once actually be learned and more importantly updated to all vehicles in the fleet.
My prediction is that by 2030, road traffic collisions involving and caused by driverless vehicles will be as rare as cyclists injuring pedestrians.
What will be the proportion of self driving to human driven, I can't see by 2030 that the numbers of AI vehicles with no human interaction will be anywhere near the majority, in fact I'm going to put a guess and say not even 5% of all vehicles on the road will be fully driverless by 2030. One of the problems is that as we see with just getting motors to have an informal speed restriction, that won't be until at least 2022 in new motors, is there's a huge reluctance to change/restriction of motorists, even by the so called friendly face of organisations like the EU road safety commission to name just one.
To take that next jump to significant numbers of AI driven motors (with zero human input) is going to take another 10 years on top of that at least, into the 2040s ... maybe.
IF within the UK we can get total motorvehicle serious injuries to that of 40 per year and 0.6 annual (at fault) deaths then that would be simply incredible. I honestly can't see that in my lifetime/next 50 years, but maybe I'll be wrong and we have a cycling/walking utopia where people on bikes don't need to be cordoned off, where people on bikes don't feel the need to cycle on the 'pavement'.
Maybe in the next 50 years we can see a drastic cut in motoring and god forbid an increase in cycle based travel, such that the need for AI motorvehicles won't be as high in physical numbers and that the programming again will have to be modified. And that's one of the main cruxes re safety, It's still all based on those that are programming and their understanding as well as governments who set the criteria (speeds, priorities etc etc)
Will the programming have enough leaway in it for vulnerable road users, for road users that are human and sometimes unpreictable? There's a heck of a lot of nuance that even the very best of the best advanced drivers don't pick up on. As far as I'm concerned as someone who has done advanced driving, has done the daily grind into and through one of the busiest cities and congested motorway systems in Europe for far too long (thank god in the past) but has more years cycling on the road amongst heavy traffic than I've been a motorist overall, I see how advanced drivers and the instruction itself fails time and again to meet the standard that would be, IMHO safe and not induce fear, alarm or distress to people on bikes particularly but also to those on foot/people just off the road.
That fear/alarm/distress IS the singular biggest factor that puts people off cycling, curently the programming thinking is based on not striking a person, that doesn't automatically mean that the vehicles presence doesn't induce those feelings which is an assault in its own right.
As i've said before, 1.5metres frm centre of wheel on a standard cycle for most people is not enough space even at a passing speed of 30mph to feel safe, it's not even enough for leeway for all types, particularly children and that will never, ever be programmed into motors because it would slow down the progress too much or require road to be widened even more. Thus even AI controlled vehicles will not be good enough to be used amongst (on the same bit of tarmac) as any vulnerable road users types at any given time.
If we truly want children of all ages and indeed adults of all ages and abilities to cycle as a means of transport then we need to banish motors from certain segments of the highway completely, I would have 2m wide single direction lanes (on either side) or 3.5m bi-directional lane with priority at all junctions along all major routes and stop up the most direct roads within urban environments or at least take one side of the highway away from motors and again have priority for cycles along with light systems at major junctions.
Segregating motors away from vulnerable road users, on say something like a 'motor-way' would be far more preferable to having AI vehicles in the near vicinity which will still present a risk and will still make people feel fear of harm even if that harm is far far less than human driven vehicles. it's a far better ultimate solution than segregating the vulnerable as we have been trying to push through. even in the 'best' country in the world, cycling is still only at 26% (significantly less than the UK in the past without any segregation) and has over 200 cycling deaths annually despite all that segregation.
Stop saying "accident" when you mean "collision"
If the software in a vehicle causes it to missclassify a pedestrian or a cyclist and then drive into them, that's not an accident, that's a collision.
There ain't no sanity clause.
I've not clicked through on the links, but I would be staggered if Google and others weren't already doing this, and in far more detail. Sanity checking both incoming data and output is fairly basic stuff in principle. What is described here doesn't sound that revolutionary.
The references only to "a video" seem particularly odd. I understand all companies working on this are using multiple sensors - normal cameras, ultrasound, radar & lidar - and combining all those inputs into essentially a 3d model of what is around the car. Objects within that model would have anticipated future movement built in.
I've seen references to google doing the equivalent of millions/billions/[insert made up number here] of hours of driving in simulations, using real world data and applying their software to it. I'd be staggered if that didn't include the sort of sanity checking described.
There was a major report by UCL for the government, that found out of 50,000 research papers on self-driving cars, only slightly more than a dozen papers each were concerned with interaction with pedestrians and interactions with cyclists.
So you may well be staggered, but they are NOT doing the research that will make them safe to use in anything like a typical congested UK High Street. There's an avalanche of money going into this - and making us want them, in advance - but there isn't the research we need going into making them safe around people walking and on bikes.
The research is here:
https://www.gov.uk/government/uploads/system/uploads/attachment_data/fil...
This is the key diagram showing how much research is being done on which topics:
You'll find 'interaction with cyclists' and 'interaction with pedestrians' are right near the bottom of the pile. That's evidently what they think of us.
The motor industry has form here: they got rid of pedestrians in the 1920s with the 'jaywalking'* myth and they'll get rid of cyclists and pedestrians again if we get crushed by self-driving cars.
*For jaywalking, see https://www.bbc.co.uk/news/magazine-26073797