[Author]

Renhong WENG, privately as independent safety and security researcher and developer, private member of China-SAE Association.

This week we discuss some safety concept in embedded software with machine learning algorithm

(0) Machine learning algorithm vs traditional software decision

(0.5) Machine Learning lifecycle

we use the tensorflow, or pytorch, or other ML model development status

Note:

• each data has itself DSAL to be risk measurement, ranking from 0~4

• Maching Learning itself can using Safety Performance Indicator to risk     measurement, abstract from UL4600

(0.9) Machine Learning Safe indicators

DSAL, SPIs

(1) single function in QM of ASIL level

(2) one function, with one safety mechanism, A or B in ASIL level

(3)one function,one function monitoring, one controller monitoring, in engine control, and like ASIL B, C

①basic concept

②Machine Learning algorithm used:

(4)Redundant cores independently to control, ASIL D compliant

①basic concept

②Machine Learning algorithmused:

Suppose in vehicle perception, we using 2 types of functionality:

i: sensor fusion between camera and mmRADAR

Sensor fusion process as following:

ii: V2X

using Lidar as input, and GPS+HDMAP for positioning, and then as well we will use the V2X generator using the V2X protocol to other cars, pedestrians, etc and then received the feedback information, then to have final judgement.

Lidar and GPS_HDMAP data input:

sensor fusion between LIDAR and GPS/GNSS+HDMAP

After ego-car andother cars or pedestrian positioning, the ego-car will bear up V2X signals,suppose here we using the QUALCOMM the PC5 interface:

And here it use the 5.9GHz band for 5GV2X services, and whole V2X software+HW+Chip will be similar as following:

So, we can get whole softwarearchitecture which one is using mmRadar+Camera, the other using GNSS/GPS+HDMAP+Lidar+Qualcomm_V2X_developer as V2X:

Above in this article, we are discussing how to get typical safety concept and system architecture in the intelligent connected vehicles, any comments please come tome.

[Reference]

1.Data Safety V3.1

2.Machine learningnotes

3.Assuring the Machine Learning Lifecycle: Desiderata, Methods, and Challenges

4.UL4600

5.Accelerating C-V2XCommercialization, from QUALCOMM