eStang

Well-Known Member
First Name
Larry
Joined
Apr 14, 2020
Threads
6
Messages
373
Reaction score
487
Location
Illinois
Vehicles
2021 Mustang Mach-E FE, 2021 Mustang Mach-E GT
Occupation
software salesperson
Country flag

noway

Well-Known Member
Joined
Jun 3, 2021
Threads
7
Messages
300
Reaction score
288
Location
Norway
Vehicles
2021 Mach-E First Edition
Occupation
Software developer
Country flag
It is probably right that there are much problems from fusing radar(or lidar) and camera based systems together. It is two different systems, and if they don't match who to trust? In computing it is always nice to have three independent systems doing the same thing, and then ignoring any input which is not verified by majority (that is two out of three). Having two systems means one system must be considered the primary and the other being the secondary. Trying to have both running actively will result in disaster if the two systems give two different outputs. It is enough to pull the emergency brake based on seing "something is wrong", but it cannot provide information about which system failed (in all cases).

The strange thing is that Mobileye was the company trying to implement pure vision AV in the first place, and Tesla did not, but Mobileye still do both, independently. Their idea is to have two systems working independently, one is vision only, the other is radar/lidar only and having both systems being capable off fully autonomous driving. The latter system will not be able to read things like traffic lights, lane markings, etc. but it will be able to drive without crashing into things with the assistance of the real time map generation (it will know the road layout without seeing it). The map could maybe in some way be considered the third system since it would be able to provide some feedback of expected result.

The pure vision system is going to be cheaper, but adding a radar or a lidar facing forward is not going to be very expensive and having this also will be able to replace the driver as a monitor. The pure vison system can drive the car while the radar/lidar system is verifying it is not doing anything wrong. If one system decides differently than the other system it could, literally, pull the emergency brake to avoid a crash (or safely stop). In current level 2 systems this monitoring, verifying correctness, is done by the driver, in a level 3 system it will have to be self-monitored, it will need to know by itself when something wrong has been done, both by itself or by other drivers.

So seeing cars having both lidar and radar based collision avoidance systems is kind of like building this monitoring system in the background and testing it, with human drivers being monitored. Humans make mistakes, and this is where these systems will prevent a collision while also providing feedback to the performance before this system would be used to monitor an autonomous driver using cameras only. I would say by seeing Tesla going for autonomous driver first and seeing others going for driver assistance first means they are just developing two different systems, but to do L3+ both systems are needed.

But probably most importantly in this context. Actually physically having one or more radars does not create sensor confusion. They could just as well have been there, but not being used actively. They still could provide feedback enough to pull the emergency brake if someone decides to paint some fake traffic on a big poster or a screen on the side or back of a truck.. They might not be an active part in driving the car, they could provide some feedback, but still be able to do important work.
 

NoMoShocks

Well-Known Member
First Name
Larry
Joined
Jul 9, 2020
Threads
19
Messages
347
Reaction score
526
Location
Camas, WA
Vehicles
2017 Chevy Bolt Premier EV, 2018 Mitsubishi Outlander GT PHEV
Occupation
Electrical product safety engineer
Country flag
But where was Sandy to tell me which way was right and which was wrong. Now I'm just left with my own thoughts....
Sandy has a different definition of Tear Down. I thought it was refreshing to have a Sandy video without Sandy.
 

EVready

Active Member
Joined
Jul 29, 2021
Threads
1
Messages
40
Reaction score
41
Location
MA
Vehicles
MB GLC300, BMW M240xi, Lexus RX350
Occupation
Retired Engineer
Country flag
Sandy has a different definition of Tear Down. I thought it was refreshing to have a Sandy video without Sandy.
What I find fascinating is that in his video today, he noted that a senior Ford engineer will be with him for a video discussing the MME cooling system. I've got to believe that after the video was posted thoroughly trashing their cooling system especially as compared to Tesla's, they had to do something - this could be interesting.
 

Excited

Well-Known Member
Joined
May 11, 2021
Threads
4
Messages
167
Reaction score
119
Location
Norway
Vehicles
Mach-E RWD SR Techno+
Occupation
GIS Engineering
Country flag
It is probably right that there are much problems from fusing radar(or lidar) and camera based systems together. It is two different systems, and if they don't match who to trust? In computing it is always nice to have three independent systems doing the same thing, and then ignoring any input which is not verified by majority (that is two out of three). Having two systems means one system must be considered the primary and the other being the secondary. Trying to have both running actively will result in disaster if the two systems give two different outputs. It is enough to pull the emergency brake based on seing "something is wrong", but it cannot provide information about which system failed (in all cases).

The strange thing is that Mobileye was the company trying to implement pure vision AV in the first place, and Tesla did not, but Mobileye still do both, independently. Their idea is to have two systems working independently, one is vision only, the other is radar/lidar only and having both systems being capable off fully autonomous driving. The latter system will not be able to read things like traffic lights, lane markings, etc. but it will be able to drive without crashing into things with the assistance of the real time map generation (it will know the road layout without seeing it). The map could maybe in some way be considered the third system since it would be able to provide some feedback of expected result.

The pure vision system is going to be cheaper, but adding a radar or a lidar facing forward is not going to be very expensive and having this also will be able to replace the driver as a monitor. The pure vison system can drive the car while the radar/lidar system is verifying it is not doing anything wrong. If one system decides differently than the other system it could, literally, pull the emergency brake to avoid a crash (or safely stop). In current level 2 systems this monitoring, verifying correctness, is done by the driver, in a level 3 system it will have to be self-monitored, it will need to know by itself when something wrong has been done, both by itself or by other drivers.

So seeing cars having both lidar and radar based collision avoidance systems is kind of like building this monitoring system in the background and testing it, with human drivers being monitored. Humans make mistakes, and this is where these systems will prevent a collision while also providing feedback to the performance before this system would be used to monitor an autonomous driver using cameras only. I would say by seeing Tesla going for autonomous driver first and seeing others going for driver assistance first means they are just developing two different systems, but to do L3+ both systems are needed.

But probably most importantly in this context. Actually physically having one or more radars does not create sensor confusion. They could just as well have been there, but not being used actively. They still could provide feedback enough to pull the emergency brake if someone decides to paint some fake traffic on a big poster or a screen on the side or back of a truck.. They might not be an active part in driving the car, they could provide some feedback, but still be able to do important work.
There are other data that could be taken into concideration as well, like weather conditions (open in numerous APIs) and how other drivers have recently responded to their cars reaction on the same spot. Not sure if Ford already is loading such variables into the system as of now …
Sponsored

 
 







Top