Continue to Site

Eng-Tips is the largest engineering community on the Internet

Intelligent Work Forums for Engineering Professionals

  • Congratulations GregLocock on being selected by the Eng-Tips community for having the most helpful posts in the forums last week. Way to Go!

Self Driving Uber Fatality - Thread III 4

Status
Not open for further replies.

drawoh

Mechanical
Oct 1, 2002
8,906
This thread is continued from thread815-437388.

--
JHG
 
Replies continue below

Recommended for you

Greg: but, the airbags must have worked very well.

dik
 
GregLocock said:
When playing with high tech toys RTFM is often advisable

“Traffic-Aware Cruise Control cannot detect all objects and may not brake/decelerate for stationary vehicles, especially in situations when you are driving over 50 mph (80 km/h) and a vehicle you are following moves out of your driving path and a stationary vehicle or object is in front of you instead.”

This puts us back to the driver gripping the wheel and watching out the front window at all times. Maybe the terminology must be clear. Forget the word robot completely. How about "observation assist"?

I wonder how a robot can deal with smoke drifting across the road? Normally, any object that intersects the robot's line of travel must be avoided. Can the robot distinguish between smoke, and a fallen tree branch or perhaps a crowd of cyclists? Once we establish that certain objects can be driven through, we create the possibility that some other object will be mistaken for smoke.

What if the smoke is from something dangerous like a forest fire?

--
JHG
 
"I wonder how a robot can deal with smoke drifting across the road? "

Depends on whether the robot can a) see it in the first place, and b) how it was programmed to deal with that use case. Barreling through smoke thick enough to blind a lidar is a foolish proposition, but few smokes can blind a radar, so a robot could be equipped to deal with smoke of various kinds. The robot could then be programmed to deal with the various smoke conditions, by coming to a complete stop, or by slowing down as necessary to maintain safety. Whether someone would be willing to pay for all the technology needed to deal with all situations is a different matter altogether.

TTFN (ta ta for now)
I can do absolutely anything. I'm an expert! faq731-376 forum1529 Entire Forum list
 
There are ways around obscurants; obviously radar works for many; range-gated cameras can get through differing levels of obscurants as well; they've been tested for landing airplanes in fog, seeing through brownout during helicopter landings, and finding submerged mines in littoral waters.

TTFN (ta ta for now)
I can do absolutely anything. I'm an expert! faq731-376 forum1529 Entire Forum list
 
Wouldn't Emergency Braking be the safety system that is expected to do the job when rapidly approaching an object like a fire truck in the cars path?
 
LionelHutz,

In my analysis in previous posts, I have assumed that ae robot must be able to come to a full stop at a rate well below that of a panic stop. If the robot decelerates at 0.4 or 0.5g, the human driving the car behind will have time to notice and stop their car, possibly in full panic mode. Remember, the human has a reaction time of at least 3/4sec. I recall from reading car magazines in the seventies, that car brakes pulled something like 0.7g. I am sure this has improved, but safety systems must be based on the lowest common denominator, in this case, an old car with worn tires. If the human does not see the obstacle the robot is reacting to, they will react to the robot's brake lights.

Assuming LiDAR is used to detect the hazard, the robot needs three scans to work out the velocity and acceleration vectors. At a 10Hz scan rate, this takes 200ms. The robot does not react instantly either.

--
JHG
 
Tesla doesn't use Lidar, they use radar and cameras and they do claim the car has an emergency braking feature.
 
LionelHutz,

I am aware than Elon Musk does not approve of LiDAR. I was not surprised when I read about this, although LiDAR looks a lot better when you do the analysis. The video cameras themselves are way cheaper than LiDAR. You need at least four vision systems on your car. With LiDAR, I assume you need one LiDAR that sweeps 360[°] around the car, one that looks forward to detect objects you are speeding towards, and two to function as your side mirrors. Somehow, the robot must detect stuff right next to the car.

I would expect the video cameras to have a better range than LiDAR, and much better resolution. Generally, you need two cameras to figure out range. I don't understand the limitations of the technology. If you have, say, ten cameras each feeding 1080p data at 30Hz to a computer, can the computer process this in real time?

My point above is that if a robot car slams on the brakes, it will be rear-ended. It needs to detect hazards in time to make a somewhat gentler stop.

--
JHG
 
I would expect the video cameras to have a better range than LiDAR, and much better resolution

Better resolution, yes, but better range only in a limited set of conditions; the cameras are a complete crapshoot without daylight. The IR illuminators in typical cameras might get you out to 25 ft, and that's optimistic. Lidars have to be designed to exceed 100 meters to do an effective job of detection and then braking in time to not hit the detected object.

I assume you need one LiDAR that sweeps 360° around the car, one that looks forward to detect objects you are speeding towards, and two to function as your side mirrors. Somehow, the robot must detect stuff right next to the car.

Not sure why your single 360-deg lidar can't do all of that.

ten cameras each feeding 1080p data at 30Hz to a computer, can the computer process this in real time?
That's just a matter of how big the computer is. One commercial satellite payload uses multiple 133Mpix cameras, all running at 30Hz frame rates, and the data is pumped into a high-end FPGA system on a chip plus GPU.

"slams on brakes"
human driven cars are likely to be rear-ended when that happens


TTFN (ta ta for now)
I can do absolutely anything. I'm an expert! faq731-376 forum1529 Entire Forum list
 
IRstuff,

I am working out here that a 360[°][ ]lidar with an 800kHz laser and 10Hz scan rate sees a spot size of 270mm at 21m[ ]range. This limits its speed to 40kph (25mph). This resolution is adequate for backing up at low speed, and for tracking large objects moving along with the robot on the highway. If the robot is going to travel at 100kph (62mph), it must resolve objects 100m in front of it. A 40[°][ ]FOV deals with roads curved gently enough to allow 100kph speeds.

An attentive human can identify a hazard hundreds of metres in front of the vehicle. At 800Hz, a LiDAR cannot possibly see anything more than 200m away. The limitation I am assuming here is that the reflection must return before the next laser shot. Yes, there are work-arounds. The robot is more likely to stop suddenly and get hit.

--
JHG
 
I wasn't posting theoretical. Tesla claims their cars have emergency braking yet it drove into the fire truck. So, why didn't the emergency braking system do anything? Maybe it did brake which wasn't enough to avoid the collision but only lessen the impact?

Isn't that the second time a Tesla hit a parked fire truck?
 
"An attentive human can identify a hazard hundreds of metres in front of the vehicle. At 800Hz, a LiDAR cannot possibly see anything more than 200m away"

No; it's more a question of how much power you are willing to throw out. The Velodyne HDL-64 has a horizontal pulse rate of 34.4 kHz, and therefore can accommodate the 2.7 us TOF for a 400-m range

TTFN (ta ta for now)
I can do absolutely anything. I'm an expert! faq731-376 forum1529 Entire Forum list
 
LionelHutz said:
I wasn't posting theoretical. Tesla claims their cars have emergency braking yet it drove into the fire truck. So, why didn't the emergency braking system do anything? Maybe it did brake which wasn't enough to avoid the collision but only lessen the impact?

Are we back here again?
 
A Tesla drove into something yet again. What other question is there to ask besides why the systems they provide on the car didn't work?
 
LionelHutz,

Tesla keeps pointing out that their cars are not autonomous. You have to watch the road and keep your hands on the wheel.

--
JHG
 
"Tesla keeps pointing out that their cars are not autonomous. You have to watch the road and keep your hands on the wheel."

True that, but it still does not explain why Tesla's Autopilot's standard function failed. I guess one might quibble about, "designed to."

[URL unfurl="true" said:
https://www.tesla.com/autopilot[/URL]]Automatic Emergency Braking

Designed to detect objects that the car may impact and applies the brakes accordingly

TTFN (ta ta for now)
I can do absolutely anything. I'm an expert! faq731-376 forum1529 Entire Forum list
 
The emergency braking is touted as a standard safety feature that is always functional, not part of the "Autopilot" or only functional when "Autopilot" is engaged.
 
Yes, that's my take on that incident. I wish my car had city stop or AEB, bizarrely of course there is no uninteresting way to find out if a given car has these systems since they are often options.

Cheers

Greg Locock


New here? Try reading these, they might help FAQ731-376
 
Status
Not open for further replies.

Part and Inventory Search

Sponsor