Agreed that the ultimate solution is infrastructure.
That's what we harp as well (Luxonis-Brandon here). And we actually see our goal as means to get there... to do what we can (as technologists) to try to keep people who ride bikes safe, keep people riding bikes, and keep there being demand for the infrastructure going.
If everyone is too afraid to ride bikes like my circle is (reminder: 1 of my friend was killed, and 3x got broken femurs, broken hips, and 1 of those got a traumatic brain injury), then there won't be demand for such infrastructure.
And with lower demand, it's harder for there to be incentive to make the infrastructure, so the infrastructure may actually get worse - making it even less safe - and then resulting in even fewer people riding bikes.
And there's another constraint: I'm an Electrical Engineer and such tech solution are all I’m good at. I don't know how to make roads with bike lanes. And more importantly, I don't know how to influence politicians to make these roads. (And my wiring is such that I'd never be good at it, no matter how hard I try.)
That, and there are ~4 million miles of road in the US. So if we put the whole economy onto making new roads as fast as we could, we'd be a decade or more realistically probably two or more decades out until we had the new roads with dedicated bike lanes.
This is what The Netherlands did, by the way, and they accomplished it (in a comparatively small country) in about 2 decades. And this accomplishment is why there exists absolutely zero car-bike accidents in the Netherlands (and much lower car-car accidents too).
So our goal is to try to help on the demand side. If there isn't demand for safe infrastructure, that infrastructure won't come. And there won't be demand if people are too afraid to ride bikes because of deaths and injuries of their friends/family.
So we want to try to use technology as a bandaid until the infrastructure is there, to try to keep people safe, and to also then help drive making infrastructure (by making demand for the infrastructure).
Speaking of which, this device quantizes risk. So when used on a bike, it would allow mapping quantized risk to bikers, which then helps to advise which are the most important routes/streets/etc. to be improved for safety. And also advises people who want to ride bikes as to which routes they should avoid.
So then when infrastructure is installed, this system can make it so the most-fatal locations get infrastructure first. So it's more prioritized and effective.
The hardware/firmware/software/AI necessary to make such a device wasn't available when we started this though, so we had to make it first. And we're releasing it with OpenCV. See here for details:
https://opencv.org/opencv-spatial-ai-competition/
So the bike-safety idea is not a product yet, but the underlying platform is now (and is going live on KickStarter soon).
Yes, totally agree on V2X. That’s another hard infrastructure problem though… as it requires so many parties to adopt it (and implement it properly) for it to really help. We did talk to a bunch of V2X folks on this though to see how we could make an integration happen… but no traction so far.
Thanks for the reassuring words sunpazed! Yes, that’s exactly right... where the Garmin product is great for alerting the rider there exists a vehicle behind them - which is SUPER useful for touring riders - this alerts the rider and the driver of an impending collision. Which is what makes it applicable to bicycle commuting, where say 500 safe passes occur per commute, and you only care about the one that’s going to hit you.
Radar works great for presence, but is extremely expensive compared to computer vision for determining exact positioning and 3D trajectory. And computer vision (specifically the combination of stereo CV/AI) is actually super-good at this, and can see quite granularly (i.e. 1 million depth points every 60 times a second out to ~20 meters).
Luxonis-Brandon here.
So the distracted driver actually wouldn't be aware that the person riding the bike has this installed... unless the driver is paying super-close attention, in which case the device is not needed.
So unlink the Tesla example, where the driver knows his vehicle has this safety feature... the driver does not know the biker has this feature.
My experience of cycling in the UK is many cyclists already go to great lengths to make themselves visible, and it makes no difference at all to close passing, etc. Hi-viz clothing, bright strobing lights - drivers act like you're not there.
It also makes drivers lazy - having nearly been knocked off my bike, the driver stopped and berated me for not wearing hi-viz, despite the bright lights I was using. Not impossible, therefore, that I also get berated for not using your device to warn me about drivers. There have been attempts to do this in the UK already with apps which notify drivers of cycists, with the app maker saying it would be the cyclists' fault if they were in an accident and weren't themselves using the app. (This was repudiated by the Advertising Standards Authority, but still.) This is similar to the all-too-frequent reporting about "A cyclist was crushed to death by a truck. Police said the victim was not wearing a helmet." Even the helmet manufacturers don't say it will stop a 20-ton vehicle from killing you :(
As Chris Boardman says about the helmet debate, it's the wrong emphasis. The issue is safety on our roads and drivers being careful around cyclists. Also, cyclists being predictable and riding safely. An electronic device will make little difference without that, and I feel it could make things worse.
Also, I find a mirror on the handlebars, a life-saver check when turning, and listening for engine noise to be pretty good at alerting me to drivers. Not sure electronics would improve this.
Thanks for the kind words. Brandon the founder here. :-)
So although I could see how that quote could appear like that, the quote is not kissing the ring. We scoured the whole semiconductor start up scene to find a chip that could be used for this.
There are only 2 chips (as I'm aware, as of this writing) that can be used in such a way:
1. Intel Movidius Myriad X
2. Inuitive NU4000AI.
And the Inuitive, until super recently (i.e. 1.5 years after we have already built hardware off the Myriad X) was not ready to be used (had tape-out issues at the fab).
So this made the Myriad X the only chip that (1) had the capabilities needed and (2) was actually available and in production.
In terms of our open-source nature, the latter is what we've implemented - where we have closed source binaries running on the Myriad X - which then have a slew of open-source counterparts on the host. Sorry if this came off disingenuous. Do you have advice on how to phrase it in a better way?
The above includes open source hardware, core capabilities in C++ which can be cross-compiled for various hosts, and open-source training notebooks and use-case examples.
So the goal of open-sourcing all that we can (including hardware) is to enable folks who have their own applications to leverage this, modify it, use it, etc. w/out having to even talk to us. We cannot open-source the code that runs on the Myriad X however, as then we wouldn't have a way to monetize, and would have to give up on the mission. (That, and we're also not allowed to.)
We agree that segregated infrastructure is the ultimate solution. But we're technologists and are 0% effective at lobbying for infrastructure. Having had a bunch of friends hurt and killed by distracted drivers, it's us using the skills and capabilities we have to try to make a difference.
And the core of the development is the computer vision technology which allows a whole slew of other applications, including other safety solutions. The sheet numbers of problems solvable with this sort of application-specific human level perception is crazy.
We'll be releasing the winners of stage 1 soon (please feel free to apply!), which will quantify how crazy it is to be able to approach human-level perception (what an object is, and where) for specific myopic tasks (like picking strawberries, or detecting when a car is going to run you over).
Anyway, back to it, I agree personally the solution is infrastructure (which I've tried to write in every location that covers this... but maybe failed to do in that interview). But I view this as a stepping stone to infrastructure. If people are too scared to ride bikes because of the injury and death at the hands of distracted people driving cars, then there will be no push for infrastructure (as why build infrastructure for a thing no one does?).
So the goal is to allow people to ride bikes -safer- before the infrastructure is there... and thereby help drive the infrastructure change (through supply and demand relationship).
Yes, ultimately infrastructure is the real solution here. Take the Netherlands for example. We interviewed folks from nearly every country on this, and the response in the Netherlands was heartwarming:
Every person we talked to had never even heard of someone being hit by a car while riding a bike (or walking). And after the interview, they interviewed a bunch of their friends/family/etc. and came back to us with those results: Still the same, not a single story of an accident.
And this all goes back to the way the Dutch design infrastructure. A ton has been written/filmed/etc about this, but here's an example: https://www.youtube.com/watch?v=XuBdf9jYj7o
Thanks for the comments usrusr! And great points. So one of the things we did early-on is to prototype the whole flow to see if you could indeed get incredibly low false-positives. The key ended up being the combination of disparity depth (which gives over 1 million depth points) and neural inference so that the system can know incredibly granularly (i.e. within inches) the edges of things and their predictive trajectories.
So at the outset of this effort (back in 2017) such a computer-vision based device that was inexpensive and embeddable was impossible. The Movidius Myriad X came out, which has all the requisite CV/AI processing to make such a thing solvable... but there was no hardware/firmware/software/AI ecosystem which allowed producing such embedded spatial AI problems.
So the `why` of this is it allows others to solve problems on embedded systems which were previously intractable. And there are a ton of them... the number of applications we've seen has blown our mind.
