LED beacon on bike to signal robot cars

'what the hell is that'

a what?

There are so many ways it could & probably will go wrong that I'm not sure why anyone is even trying to create a standard at this time.

NT

There's no end of unusual appearing combinations that appear on roads that include humans, animals or objects of value or weight/hardness. To fail to recognise them all as something to avoid is a pretty basic screw up. Did so meone really design a driving computer without doing a fault tree? Or fail to include someone moving something across a road?

NT

Oh. Kinda like the internet of gizmos, gadgets, and contraptions?

Hmmm... whatever happened to the smart highway? V2H = Vehicle to the highway

QTH is ham radio lingo for your station location.

To ossify the technology into what can be achieved by the early implementers and adopters. It would not do to have their investments go down the drain if someone arrives late with a better idea.

More like

Any first generation computing technology is bound to become obsolete fast. It won't protect anyone's investments. It's a project bound to fail.

NT

Ok, point taken. It might be a problem if those are road legal. In my never humble opinion, none of those pictured will be a problem if they present themselves as a large object on the LIDAR. It's the small objects like kids, bicycles, scooters, Segways, and soccer balls bouncing down the street, that are the more likely problem.

I wonder how it works with camouflage vehicles, which are difficult to distinguish against the background?

In my checkered past, I worked with a few committees attempting to hammer out a standard. I was not thrilled with what I saw. Basically, the various company representatives were there first and foremost to protect the patent portfolio and early products of their employers. For example, in order to get TI to vote in favor of the 802.11g standard, TI's unique and inferior implementation was included in the standard. Same with frequency hopping instead of direct sequence. So, instead of a single standard, there were three. Same with 802.11n which had to include beam steering, beam forming, as well as MIMO, which was the original purpose of the standard.

and if that video is indicative of how well the car can"see" in the dark, it should not be allowed on the road at night.

mark

I am sure the lidar sees a lot better but the mystery is why the cars decision making software did not correctly identify a solid obstacle in its path and apply the brakes even if it couldn't stop in time.

The standard committee should add the LED beacon detector to robocars. Like a TV remote control, a 50 kHz flashing IR LED can be detected in the presence of many problems. Rich people will put bright LEDs with digital messages, poor people will have dim bulbs and get smashed. Rich folks will be standards compliant on their bicycles, illuminating cars with verbose mode insurance. Poor bikers will be given free beacons that just flash at 50kHz. Given by the robot designers who practice safe specs.

I think you're right but they didn't ask me.

But 5G will do *everything*. Promise! g

It's called "V2I" or "Vehicle to Infrastructure", covered under V2X, really.

And so someone will spend the money to do something. Without standards, no one would invest the money.

The goal has to be not to perform at least as well as a human driver, but t o perform *significantly better* than a human driver. As someone mentioned , the camera does not see as well as a human in the same light range. I th ink I would have seen her in plenty of time to stop. When you say she step ped out of the "shadows", there weren't any shadows, rather the limited ran ge of the camera meant she couldn't be seen until she was in front of the h eadlights. From a wired article, "Many experts say the car should have pic ked up on a pedestrian in a wide-open roadway, after she had already crosse d a wide-open lane." No "shadows" involved.

I would also point out that I read this system uses IR cameras to view the road, "Uber?s self-driving prototypes use sixty-four laser beams, a long with other sensors, to construct their internal map". There is no rea son why a person should not be visible in or approaching the roadway.

It is hard to tell really because the camera doesn't see as well as a human . Also the camera doesn't have peripheral vision. People do.

Indeed.

Why is this a challenge? I have also almost hit things or the other night I was coming home and saw headlights approaching as I stopped before making a left turn. The headlights appeared to be a distant car until just befor e I was about to start my turn a car whizzed past in the other lane. It ha d a headlight out and there was a car further behind it that made the two t ogether look like a single car. People make mistakes like this all the tim e, around 35,000 times a year that kill someone. Machines can and will hav e to do much better before we trust them, but that isn't very far away.

Rick C.

So long as the AI is no worse than the average human driver it poses no more threat to the rest of us than any other driver on the road.

It will need examination of the sensor logs to figure this out.

There was certainly clear line of sight on the pedestrian in the road but she did step in front of the car with under 2s to impact. A human driver using the UK approved stopping distances would have barely enough time to hit the brakes before they hit her if the first inkling of a problem was first sight of the bike wheel hub.

The message is clear - if pedestrians/cyclists step out into the path of vehicles at night wearing black and without lights they can get killed.

Peripheral vision wouldn't help here the problem was in the forward visual field. The camera images shown lack dynamic range. I presume the car has other lidar based data but that didn't trigger braking.

The challenge is that pedestrians and cyclists offer a very small cross section to be seen and if they are travelling without lights and wearing ill advised dark or road coloured clothing practically invisible.

Machines have to do at least as well as humans. But then you have some very awkward ethical problems for self driving cars. Is it for instance acceptable to kill an entire bus queue of people to save the life of one occupant of a single car if that is what the AI determines is needed?

Too expensive. My guess(tm) is that bottom of the line technology would be a retro-reflective patch, with a color and pattern that is easy for the LIDAR system to recognize. Poor bikers will be issued patches that identify the wearer as a cyclists, pedestrian, vehicle, brick wall, homeless, emergency vehicle, etc. High tech versions will include an RF transponder, similar to what is currently in use in airplanes, that detects a LIDAR scan, and responds with a burst of RF.

Oddly, I worked on early versions of both of these technologies in the distant past. About 1975 methinks. Before cheap GPS, the transponder was a "signpost" system, where small low power transmitters were to be installed in traffic lights and light poles. When an subscribers vehicle drove past one of these transmitters, a receiver in the vehicle would record the transmitted position, and retransmit it on a different frequency to the company or agency dispatch center. I was initially used in large open pit mines to keep track of where the big ore trucks were located.

The LIDAR transponder manifested itself in what I called a "portable telephone ringer". At the time (about 1981), I was working in a large in a large open building with high ceilings and a few partitions. Packed into this space were about 150 office desks and probably as many telephones. The phones were ringing constantly, making it difficult for me to think and work. I decided that the world needed a better way, which was to have the phone "ring" to an IR transmitter located near the ceiling. Everyone wore ID badges to which I added an IR filter, photodetector, upwards facing optics, some electronics on the back, and a beeper. Instead of the phone ringing, the ID badge would ring. One would then go to any phone, dial their own extension, and pickup the call. I had prototypes make and patent applications prepared but delayed the idea long enough for cheap voice mail and cordless phones to render the idea as useless. Still, I think it might be useful to issue LIDAR optical transponders that work on the same principle.

I can only comment on the RF part of the puzzle. 5G might arrive in various flavors. So far, the predicted, likely, and allocated frequencies are:

USA 3100 - 3550 MHz, 3700 - 4200 MHz Europe 3400 - 3800 MHz China 3300 - 3600 MHz, 4400 - 4500 MHz, 4800 - 4990 MHz Japan 3600 - 4200 MHz, 4400 - 4900 MHz Korea 3400 - 3700 MHz

USA: 27.5 - 28.35 GHz, 37 - 40 GHz EU: 24.25 - 27.5 GHz Korea: 26.5 - 29.5 GHz Japan: 27.5 - 28.28 GHz China: 24.25 - 27.5 GHz, 37 - 43.5 GHz Sweden: 26.5 - 27.5 GHz

All of these allocations are either speculative, tentative, and preliminary. The technology needed to make things happen in these two frequency ranges is quite different. 3 and 4Ghz can probably be convinced to work for 5G because the technology is quite similar to existing cellular systems. 30GHz is quite different, requiring beam steering and/or directional antennas at both end. Even with elaborate antenna technology, range is limited at these frequencies. The solution is called "5G densification": which means the cell sites will be more numerous and closer together. More detail if anyone wants it.

Everything built on top of the RF part of the puzzle will probably be just fine, if the RF works as intended. With luck, it might even scale to large systems as the models predict. If not, 5G could still do everything promised, just not very well (Mediocrity for the Masses).

Some detail on the Velodyne LIDAR system used by Uber: (docs near bottom of page) Note the sample photos of what the LIDAR sees. It should have seen the pedestrian. So, where are the LIDAR reconstructed videos from the accident?

If the LIDAR system didn't see anything, there is one potential problem with using IR. If the clothes is IR absorptive, there will be no reflection. Worst case, she might have been wearing something close to Vantablack: "Vantablack absorbs 99.965% of visible light" Also works at IR frequencies: It's also possible that she was wearing something that only absorbed specific IR frequencies, such as whatever is used by the LIDAR.