Is Seeing Machines putting its human-perception technology onto Ambarella’s robotics processors?

Seeing Machines has a relationship with Ambarella that is considerably more substantial than a conventional technology partnership. The two companies first announced their collaboration in January 2022, when Seeing Machines agreed to bring its driver and occupant-monitoring technology onto Ambarella’s CV2x family of CVflow AI processors. The arrangement allowed Seeing Machines’ embedded Driver Monitoring Engine (e-DME) to use Ambarella’s CVflow acceleration engine, combining SEE’s perception software with Ambarella’s edge-AI silicon.

The relationship subsequently produced a concrete, integrated solution. In 2023, Ambarella, Seeing Machines and Autobrains demonstrated a single-SoC system combining forward-facing ADAS, driver monitoring and occupant monitoring. 

Crucially, Ambarella described its CVflow platform as an “open” AI-processing architecture that allowed both Seeing Machines and Autobrains to port their software stacks onto its silicon. 

This matters because Ambarella has subsequently expanded well beyond automotive. Its latest CVflow portfolio is explicitly targeted at industrial robotics and physical AI, with its CV7 SoC supporting applications including robots, drones, ADAS and DMS. Ambarella says its common CVflow architecture and SDK allow perception software and AI models to move across generations of its silicon with relatively little redevelopment. 

At precisely the same time, Seeing Machines is attempting to make a similar transition. Its new Physical AI platform and Perception Map technology are designed to give machines a real-time understanding of people, objects and their spatial relationships. The company has already announced a robotics proof-of-concept with a global industrial technology company focused on factory automation and human-robot interaction.

There is currently no public evidence that the new Seeing Machines robotics platform is running on Ambarella CV7 or that the unnamed robotics customer is using Ambarella silicon. It would therefore be premature to claim that SEE’s robotics technology is already part of Ambarella’s robotics offering.

But the possibility is compelling.

Ambarella is positioning CVflow as the low-power perception engine for Physical AI: cameras and sensors feed into its chips, which process the environment in real time for autonomous machines and robots.

Seeing Machines, meanwhile, is developing the specialist software required to understand the human within that environment — where they are, what they are doing and potentially what they are about to do.

The automotive relationship therefore provides an intriguing precedent. If SEE’s human-perception technology can be ported from Ambarella’s automotive chips to its latest robotics processors, the companies could potentially offer a powerful combination: Ambarella providing the edge-AI compute and Seeing Machines providing the human-perception intelligence.

That would move SEE’s opportunity well beyond driver monitoring and into the rapidly developing market for embodied or Physical AI.

Conclusion

When SEE demonstrates that its perception stack can be ported, the opportunity could extend well beyond Ambarella. Other chipmakers — from NVIDIA and Qualcomm to NXP and Renesas — are all competing to provide the computing platforms that will power the next generation of robots. 

If SEE can provide the specialist human-perception layer that sits on top of those platforms, chipmakers could have a strong incentive to incorporate or support its technology rather than develop equivalent capabilities themselves.

And that could fundamentally change the economics of the opportunity. Instead of Seeing Machines having to win every robot manufacturer individually, its technology could potentially become part of the reference platforms offered by the chipmakers themselves — giving robot manufacturers access to SEE’s human-perception capabilities as part of the underlying AI platform. 

For a company that has spent decades developing technology that enables machines to understand humans, that could open a market far larger than driver monitoring. Even a modest software or IP fee per robot could become significant if SEE’s technology were deployed across millions of machines.

The potential market

Goldman Sachs estimates that the humanoid robot market could reach about $38bn by 2035, with a more bullish scenario reaching $154bn. Annual humanoid shipments could rise from more than 250,000 in 2030 to 1.4 million by 2035.

Humanoids, however, are only part of the opportunity relevant to SEE. Its Physical AI platform is aimed at industrial automation, robotics and human-machine interaction, rather than humanoid robots alone. If SEE can establish its technology as a portable human-perception layer across multiple chip platforms, even a relatively small slice of this emerging market could become meaningful.

The question for management at Seeing Machines is therefore: “Is the new Physical AI/Perception Map platform processor-agnostic, and has it been ported to Ambarella CVflow — particularly CV7 — as part of the current robotics POC?”

The writer holds stock in Seeing Machines.