Valeo: See the Invisible, page-274

Hi rob,

1. The Valeo presentation was about seeing the invisible.

2. They spoke about developing their Lidar IP in-house.

3. Their patent for their hybrid camera system combines data from a standard camera with data from a DVS/Lidar.

4. The patent talks about using NN to analyse the normal camera data.

5. The patent does not mention NN in connexion with Lidar data.

6. The patent dates from 2018.

7. BrainChip began working woith Valeo mid-2019.

8. Subsequent to the "partnership", LDN mentioned Akida as having a sweet spot for Lidar.

9. Lidar projects a pulsed narrow scanned light beam over the field of view and and detects the time of the returned reflection via a lens and shutterless photodetector (which we wil call DVS).

10. LIdar is designed to detect changes in pixels.

11. Lidar on its own is not useful in mobile applications where the DVS and laser are in motion without massive compute resources and time delay because the relative position of everything in the field of view and the reflected light to the associated pixels changes with the DVS camera position.

12. The wizardry of Valeo's patent is that, by combining the camera pixel data with the DVS data by effectively subtracting the camera data from the DVS data, any moving objects in the field of view are indicated by the remaining Lidar data after the camera data is subtracted. I consider that this is at least part of the Valeo IP mentioned in their presentation.

13. The resulting DVS output is in a SNN compatible format so it can be fed directly to Akida, hence Akida's sweet spot for Lidar.

14. However, I am not aware of any Valeo patents which cover the NPUs (network pocessing units) which make up an SNN.







Valeo do have another Lidar patent:
DE102019115327A1 Linienmarkierungsidentifizierung unter Verwendung von LiDAR

which uses olde worlde CNN to identify line markings and is apparently only filed in Germany.

This dates from mid-2019

I think that there is a good possibility that Valeo are using Akida IP in their Lidar sensor, but I may be wrong - first time for everything.