BrainChip Demonstrates Smart Automotive, page-67

I can't remember who posted but I think someone previously suggested the customer for this smart automotive video may have been Valeo or Renesas.

My view is that it's probably none of them. Renesas has signed an IP license for 2-nodes (8 NPU's). This IP license is for a situation where latency is not an issue, yet in the example above it requires almost an entire chip (71 out of 80 NPUs). It is possible that they are working on multiple products which haven't been disclosed yet, but I'm not convinced. The Brainchip example is a higher end system than required for most everyday cars so overkill for most applications Renesas would be considering IMO.

Valeo on the other hand is working on a POC for a lidar system. So I would guess this is for their next gen SCALA systems (SCALA 3).

It's possible that it's Ford, though they would have used their own car here.

Alternatively it's another undisclosed car company, potentially using a Socionext radar. Socionext has demonstrated videos showing several different types of in-car detection, though only been with radar. This ranges from things like detection of the number of occupants in the car, burglary detection. However, Socionext's test car in their videos is a Toyota, and in the Brainchip example I think someone on this forum mentioned it was a BMW. As BarrelSitter mentioned, the radar seems to be built into the rear-vision mirror, which is the same setup Socionext uses (see screenshots below). As they are one of Brainchip's customers, it's reasonable for Socionext's radar is being used for this demonstration.

It's also worth noting the radar seems to be activated first in the video, probably because it is very low power. If you read the specs on the Socionext SC123x series, it draws 0.5mW average power for human detection systems with low duty cycle (see link below). It looks like the microphone might be always on, but again also is also a low power application. However, it looks like the face identification system is only activated once the radar detects someone (watch the face identification system - it is greyed out before the radar detects someone and then turns on) to keep power consumption as low as possible. This to me indicates it has been setup for a customer, as otherwise they could just have each system running constantly at all times. It takes more time to set it up and test this as opposed to having all systems on at all times. In addition, the car has been setup reasonably neatly (there aren't messy cables dangling around). This indicates that either Brainchip are going to extra lengths to deliver more professional videos as they approach commercialisation or this is an actual customer setup.

One thing I'm not sure on is why both Aaron and Todd touch the rear view mirror when they get in the vehicle. The mirror is aligned directly at their faces so it doesn't appear to need adjustment. And touching the mirror doesn't appear to impact on the system behaviour. All of the detection systems seem to be online and working before they get in the car. The only thing I can think of is maybe they are pushing a button to activate the visual wake word system just before they speak. It's possible it was being triggered by other noises so they set to only activate on button press.

Regarding the personalisation aspects, these are more like what I would expect from Ford. They are working to develop a system that recognises the driver and adjusts mirrors, seat, audio etc. To do this I would anticipate them using sensor fusion with multiple sensor types (such as vision and audio) to minimise the risk of the false identification. However, why would Ford go to all this effort and not deploy it in their own car? They would want to confirm the layout of all the components so they could figure out if anything needs to be rearranged in a production vehicle. This indicates it's probably not Ford.

In one of the recent Brainchip presentations one of the presenters indicated the cost of an entire chip was expensive for many of their customer applications, so it's likely many will use smaller quantities of IP. In the Brainchip video example they are roughly using an entire chip, which is relatively expensive. I wouldn't think the cost of a system like this would bother Ford, as they just seem keen to be seen as leaders in such technology. IMO they would be more concerned with the system performance and reliability over cost. As JDelekto previously mentioned the setup used by Brainchip was more snappy than their previous demonstrations. I'd argue it had a faster response than is needed for this use case in a commercial situation. For an average car maker I would suspect it could be twice as slow and still acceptable. So it's possible this video was purely to demonstrate Brainchip's performance capabilities. Alternatively this example points to someone prepared to pay a premium to demonstrate top of the class performance (potentially Ford), or someone with a luxury brand (such as BMW, whose badge was on the steering wheel). There's a good chance the use of a BMW was not accidental here.

Pure speculation, DYOR

Socionext in cabin radar use cases:

Socionext radar system in rear view mirror:


https://socionextus.com/pressreleases/application-note-highlighting-radar-for-optimized-sensing-accuracy/

24GHz SC123x Series

Measuring at a compact 9 x 9 x 1mm, the Socionext SC123x series all-in-one 24GHz radio-controlled FMCW (frequency-modulated continuous-wave) radar sensors feature wide detection area and are well-suited for battery operation with an ultra-low 0.5mW average power requirement for human detecting applications (0.1% duty cycle). The 24GHz sensor perceives entry motion and adds distance detection without the need for an external MCU for performing calculations. When paired with an MCU, the sensor can also perform 2D angle detection.