AI bike light detects cars, alerts bikers to bad drivers
 

Raspberry Pi-powered AI bike light detects cars, alerts bikers to bad drivers
Data from multiple Copilot devices could be used for road safety improvements.

Link: Ars Technica article


It looks a bit clunky and is quite expensive, but as a prototype, it sounds promising ....

​​​​​​​

Interesting idea. It makes sense that as we're beginning to see self driving tech for cars that related tech would become available for bikes. I do believe sometime in the future (years, decades from now?) pretty much all things related to transportation will interact to improve traffic flow and safety.

While there will certainly be some hiccups, I expect cyclists will ultimately be a lot safer with self driving cars. And tech like this for bikes would just make it that much easier. Cars see bikes, bikes see cars and they can all communicate and share information.

the tech starts off with good intentions, but soon enough it'll be exploited for a bad cause.

I wanted to hate this as another example of AI being incorporated into something that doesn't need it simply because it's the new hot thing... but honestly this looks pretty useful.

Inevitable, yes. Also inevitable that the 'hiccups' along the way are going to kill people in cars, on bikes, on foot, etc., and it's going to be very hard to litigate fault.

Actually the tech starts out with exploitation (of y'all's fears) as the basis. It can't do any of what it says. Not that it has to. Truly bad stuff happens so rarely (despite what y'all's think) that the Copilot will never actually have to prove itself. WHEN it fails to protect you ... oh well ... in the best case (for them) no one will ever know, and in the worst case, the rider will be blamed for being unable to respond to the warning. And mainly that is the case! If a cager wants to off you, then you are off. There is no existing technology that can save you. Save your money instead.

Even if a cager doesn't necessarily want to off you ... well, not that long ago, in my city, a cager in a pickup (really) with bald tires (in the rain) tromped on the gas when the light turned green and spun out instantly whacking the cyclist next to him into oncoming traffic. RIP. Happened in a split second. With a Copilot, the cyclists crumpled bike would have a $400 accessory that will be salvaged with the rest of the kit. The outcome would be the same. I don't think I have ever told the BF readership what to, or not to, do with their hard earned (or not) money, but I hope no one in this thread buys one of these things. Actually, I'm hearing some pretty disappointing things about the Varia products. I could easily buy one of those, but just wasn't entirely convinced as to their worth. I'm glad I held off. None of this stuff is ready for prime time. But $400 for a souped up Cycliq is just beyond the pale. Don't do it.

By the way ... AI? Do you know how much computer power real AI requires? Do you think Chat GPT can run on anything less than two mainframe computers in a RAID array? Chat GPT only parses datasets of written language. The Co-pilot is (presumably) making complicated analysis of object motions in real time. You'd need a minivan to haul around the computer that could actually manage the data that a real AI system would need for this use case.

Really? I find my Varia extremely useful while riding on narrow country roads with occasional high speed traffic. Most of our local club riders are now using them and I have not heard of a single complaint. I certainly don’t have any. It’s one of the best things I’ve ever bought for traffic awareness while cycling. The rear light is great too.

Where I ride, most cars are going 10 to 30 feet per second faster than I am. So when that thing beeps to warn me of a "bad" car that might run me over, I doubt I'd have time to react.

I suppose I should be in fear of any car that less than 30 feet away from me.

I'm all for technology, but this devices current thing isn't going to help. I use Varia Radar too, and I'm of the same opinion, it's not going to alert me in time to avoid a car that intends to do me harm.

At best it can only make me aware of them. But in most cases, I know they are coming from the road noise made by the car just a split moment before my Varia alerts me.

The Varia works great for me.

The only shortcoming is it doesn't (or cannot) distinguish between my wife closing in from behind on her e-bike, a safe car driver, and a psycho in an FUV. If the AI aspect of this prototype can differentiate between those, it is a significant advance that I would pay for. The idea that a rear camera would record close or threatening vehicles preferentially is an added bonus. Garmin could license the technology for a future Varia, or develop their own version.

AI tech was never made with good intentions, imo. It’s all complete garbage and an affront to humanity.

It isn't AI being incorporated into something that doesn't need it. It's more or less routine functioning of computer technology being called AI for marketing purposes. There is no AI ... yet. The world does not yet have true AI technology. We aren't even close! What is being called AI is just fast processing of existing data sets by processers operating in the single digit gigahertz but to our 3Mhz brains (albeit CONSCIOUS brains), the computers seem 'intelligent'.

It does not take AI to color an image. Photoshop has been doing that for years. Deep fakes have been made for years. The ability to offload the processing of image data to a mainframe somewhere have it do whatever processing is necessary and send the results back to you over the internet is what makes things like ChatGPT possible. America still uses the T1 Internet bandwidth standard. Europe and Asia use the E1 standard. It is VASTLY faster. Unimaginably faster. You still would need a way to communicate with the server and the Copilot has absolutely no way to remote process the data it receives from its camera and believe me, or not, a Rasberry Pi is simply not up to any real predictive analysis of the behavior of vehicles behind you.

Even if it was, human reflexes are simply not fast enough to act on the information. Collision avoidance systems in aircraft alert the pilots to the emergency as a courtesy! Without any AI involved, computers in each closing aircraft coordinate a response such that the aircraft move away from each other. At a closing speed of Mach 1.5 there is no time to correct for both aircraft altering their paths in a way that keeps them on a collision course.

The collision avoidance systems in cars that alert you if you are drifting out of your lane do not require AI. Neither do the ones that guide your car back into its lane without your involvement. When, and if, Copilot like devices can actually DO SOMETHING about the overtaking car on a collision course (apply its brakes!) then a consumer has a device worth paying $400 for. As it stands. Someone is out to make lots of money at our (the cycling consumer) expense.

It was in another thread I saw someone mention being less than chuffed with their Varia. But I wanted to bring some links to this post so I Googled 'Varia shortcomings'. You should do the same. There are simply too many from which to choose. I don't own one. I've never even seen one. I have no idea. I'm just putting it out there that not everyone has found them problem free.

The AI here is presumably some form of machine learning which uses initial data on a vehicle's motion to predict later behavior and does some other tricks. That's a very basic, valid, and straightforward use case, aside from whether it's of any real-life value to a cyclist (questionable). It's also not clear from the description whether there's any "AI" on board or whether the gizmo simply incorporates an algorithm developed on some training set of recordings in a lab somewhere.

Yeah we can all Google for the result that we would like to support our view. Google rarely disappoints in that respect.

I’ve been using a Varia 515 paired to an Edge 530 for about 2 years now and wouldn’t ride on the road without it. I have had no problems with it and find the approaching vehicle warnings invaluable. It’s one of the few cycling gadgets that has exceeded my expectations. But feel free to convince yourself they are not worth having on your bike.

As soon as I read "Raspberry Pi powered" my first thought was, "How bulky is this thing?" The pic in the article confirmed my suspicion.

It's too bad Garmin didn't make this first. They could have called the product VarAI.

My hope is that they will license this or develop their own analogous technology and fit it into the current Varia casing.

Same here, although it is my second Varia. (I got gen 1 when it first appeared.)

The kid just got a new Canyon Grizzl, so we just got him a Varia 515 today. It is well worth every penny.