Of the myriad AI technologies and concepts exciting folks in 2026, physical AI is up there. PwC, in a strategic guide published to help ‘capture value as AI enters the real world’, forecasts the global physical AI market to hit more than $500 billion (approximately €430bn) by 2030.

The rise in capability of use cases, from robotics to autonomous vehicles, continues apace. Underneath the hood, however, is edge AI, running AI on-device so physical AI operation can be possible.

Edge Impulse has long been a leader in the edge AI space, and as a senior solutions engineer, Heath Blandford spends his time helping customers and prospects understand these concepts not just from a technical perspective, but from a problem-solving point of view. Blandford says he is blown away by the pace of change.

“The speed has been just incredible, the shift in the velocity at which a lot of these companies are moving,” he says. “I think that edge AI, and moving into physical AI as these two worlds meet… we’re really past that trough of despair where these solutions are no longer a ‘will they or won’t they’. It’s much more of an ‘it’s here, it’s ready’.

“We’re seeing things move, really, from a space where things were on the bench in R&D facilities, where teams are just trying to figure out if it’s possible, to ‘we’re going from that prototype’, right out to delivering products that are ready to be adopted, which has been really exceptional,” Blandford adds.

There has similarly been an understanding that the rapid pace of edge AI development will have a major impact on the embedded industry. As power and performance are among the primary drivers for moving AI to the edge, along with security, so that dovetails with two key considerations for embedded system design.

Blandford notes that, amid this, it is very easy to get carried away and think bigger than you should when designing for projects. For embedded systems, this is not that easy.

“A big thing that I’m running into [is] where we’ve got these lofty North Star-type projects where, really, the most high-value, high-impact thing can be in a simple solution,” he says. “Embedded systems are expanding into a bunch of different areas, but because embedded systems are what they are, there’s not really a lot of wiggle room for being able to move fast and break things. So we just have to balance that innovation speed with that long-term reliability and compliance.”

In a way, it is an in-built advantage. “As an engineer, constraints are a good thing in some cases, because they make you come up with creative solutions around those constraints,” says Blandford. “And an embedded system is exactly that… it is a highly constrained piece of hardware, and that requires creative engineering solutions, and that takes time. It all helps itself in a kind of circular way.”

A tangible example of getting this right can be seen in the work Edge Impulse has done with HP Poly. For the company’s Poly Voyager Free 60 earbuds, the goal was to add voice commands so the user can answer or ignore an incoming call.

The benefit of running the keyword-spotting detection algorithm on-device is evident, while being able to process human audio on-device was also beneficial from a security perspective. The data collection and labelling is also taken care of. Employees could, through a web-based tool, click a link and record 30 seconds of their voice for the training set, while there was additional benefit in iterating quickly on the data set, as well as the machine learning algorithm itself.

Was it a fully-fledged use case out of the box, or did it need to be finessed? “It was a little bit of both,” says Blandford. “[HP] knew they wanted to improve the experience, and they knew that probably the best way to do that was running something on-device, but there was a consultative approach to work together to find the best solution.”

Part of the evolution with these projects is in getting the broader stakeholders involved. In conversations with most prospects now, Blandford notes, it is not just the embedded team he is speaking to. “Teams are being formed now to get these use cases to market with the understanding that they will need a little bit of ML engineering, a little bit of embedded engineering, a little bit of software,” he says. “And all of these things have to come together to solve these problems.”

Looking further afield however, Blandford sees the next big step as cross-industry standardisation and interoperability. As the importance of safety-critical embedded applications grows across industry, Blandford sees a move more into the space of being able to ‘go through the same functional safety reviews or certifications… to be able to deploy maybe a single idea or single use case to multiple different industries and be certified for safety and compliance.’

Analogous to this is standardisation across interfaces – and this is something Blandford will be exploring in his speaking session at Microelectronics US, on April 22-23 in Austin.

“If you’re a hardware vendor, or a software platform, or a systems integrator, it’s very difficult to work with all of those different interfaces,” says Blandford. “So being able to come together on a single common interface for things like data ingestion, data pipelines, retraining, models, redeployments, orchestration, fleet management, being able to maybe not standardise but have some best practices around those common interfaces, can really help build that synchronous, harmonious relationship between industries.”