Brainchip is in a position of advantage simply because it is a hardware start-up with innovation in neural networks.
The car manufacturers are writing on board binaries to make things happen. At uni we worked on reverse parking (find and park) for Lexus which was done via a propriety software. The total cost of the unit was $900.
BRN offers a no software solution. Just from the top of my head I can think of few inherent advantages
- Minimal lag in decision making
- Low power consumption
- Exponential rise (advantage) in time to action
- Decision trees are independent of the architecture of the software
If you look at the current state of the google driving cars various inputs are fed onto a CPU which are then parsed by libraries calculating various angles for collision avoidance. You will be surprised that Google actually uses a neural network for their on car processing. Imagine if the car was fed with large neural network just for collision avoidance without having to pay the cost of lag and power. Electric cars stand to gain a lot.
Here is what Google's CEO Sundar Pichai has said recently
"The future is machine learning and and extension of artificial learning that build systems. The one used in Google self driving cars. These cars recognize other vehicles have Artificial intelligence and deep neural networks. They are incredibly powerful," said Pichai at an interaction with students of Shriram College of Commerce in Delhi. " We are barely scratching the surface. You can do in healthcare.. There is a lot left to do".
When the world largest technology company which has dwelved into a lot of innovation talks about utilizing neural networks for their cars one must pay close attention. Listening to hotcopper members who think neural networks on hardware is a novelty & makes a lot less business sense are probably people who mill make their fortune critiquing others work their entire life.
So to answer your question BRN offers a solution where car makers do not have to deal with an onboard software and binaries to be fed to their processors for execution. This job could just be done via training the neural network and utilizing that neural network to perform the intended action at a time rate which is exponentially lower than the one consumed by software.
(20min delay)