Understanding Triterras Current and Future Valuation, page-77

Given the 18 month blackout on patent applications, we would not expect to see anything produced by the BrainChip/Valeo joint development before December 2021 at the very earliest.

This is a Valeo patent for ultrasonic detection with CNN from 2019 (pre-joint-development with BrainChip).

DE102019107126B3
https://worldwide.espacenet.com/patent/search/family/071739314/publication/DE102019107126B3?q=DE102019107126B3

The invention relates to the processing of an ultrasonic signal by means of an artificial neural network. In order to improve the performance and to increase the accuracy of the detection of the obstacle (2), a method for processing an ultrasonic signal with the following method steps is proposed: c) Using a discrete envelope curve (6) of an ultrasonic signal with N samples as the input position (7) a first artificial neural network (10), d) generating a window extraction layer (12) of the first artificial neural network by dividing the input layer (7) into several windows (8) with a predetermined window width F, each of which is shifted by a predetermined increment S , and applying a convolution operation to the windows (8) of the input layer (7) by means of a number k of convolution kernels (9) with k> 1 by performing a respective convolution operation within each of the windows (8) for each convolution kernel (9), so that in the window extraction layer (12) for each window (8) a signal component with one value for each convolutional curve rnel (9) is generated according to a respective convolution operation, so that the window extraction layer (12) has a total of T signal components with k values each, e) using the T signal components of the window extraction layer (12) with the k values individually as the input layer for a second artificial neural network (11).


They also use Lidar for line marking detection.
https://worldwide.espacenet.com/patent/search/family/073459622/publication/DE102019115327A1?q=DE102019115327A1

The present invention relates to a method (10) for identifying line markings (24) on a road, the line markings (24) defining the boundaries of lanes, using a LiDAR-based environment sensor (14) for use in a driving support system (12) of a Vehicle (10), with the steps of providing a deep convolutional neural network (26) with an input layer (28) and an output layer (30), receiving sensor information from the LiDAR-based environment sensor (14) which covers a road section in front of the vehicle ( 10), supplying the sensor information from the LiDAR-based environment sensor (14) to the input layer (28) of the deep convolutional neural network (26), processing the sensor information supplied to the input layer (28) by the deep convolutional neural network (26) and receiving Identification information (42) of the line markings (24) from the output layer (30) of the deep convolutional neural network (26).The present invention also relates to a driving support system (12) for a vehicle (10) with a LiDAR-based environment sensor (14) and a processing unit (16) connected to the LiDAR-based environment sensor (14) for receiving sensor information from the environment sensor Driving support system (12) is set up to carry out the above method.


This one's for point cloud (ball on road problem):
https://worldwide.espacenet.com/patent/search/family/072943110/publication/DE102019111608A1?q=DE102019111608A1

The invention relates to a method for determining a proper movement (E) of a motor vehicle (1) by means of an electronic vehicle guidance system (2) with the following steps: - generating a point cloud (12) of an environment (5) by means of an electronic computing device (3) of the electronic vehicle guidance system (2) by capturing the surroundings (5) by means of a capturing device (4); - First evaluation of the point cloud (12) and second evaluation of the point cloud (12) by means of the electronic computing device (3); - Determination of a dynamic object (6) in the environment (5) by means of the first evaluation by means of a first neural network (15) and / or by means of the second evaluation by means of a second neural network (16); and - determining the proper movement (E) by removing the dynamic object (6) on the basis of a point cloud (33) reduced by the dynamic object (6). The invention also relates to a computer program product, an electronic computing device (3) and an electronic vehicle control system (2).

They also have a clunky DNN control system:
DE102018119912A1

It seems they have all been filed only in Germany?

They have a lot of applications which would benefit from Akida.

I suppose they will get IP licences, because most of them would not justify going full Akida. On the other hand, it may be quicker and cheaper to get the Akida SoC.