TG0 is making materials intelligent with low-power AI

We’ve become used to seeing robots dance, sort packages, run very fast half marathons and play amusing football matches. But the idea of having a robot in your home, unloading your dishwasher or cooking dinner still feels alien.

It might be more appealing to let robots into our homes if there wasn’t a very real risk that they could accidentally harm us or break things just going about their daily tasks. This is partly because the AI systems that drive robots are still relatively basic. But it’s also down to their lack of a sophisticated sense of touch – one that humans take for granted.

This is one of the problems that Dr Liucheng Guo, CTO and co-founder of TG0, is working to solve. Guo is recognised with a 2026 Princess Royal Silver Medal by the Royal Academy of Engineering for the embedded “AI for sensing” approach that TG0 has developed to transform physical ordinary materials into intelligent ones. The first-in-kind technology could – as well as equipping robots with human-like ‘skin’ – make our interactions with everyday technologies such as cars more intuitive. 

Guo describes TG0's approach as 'AI for sensing', using lightweight embedded AI to give physical materials a sense of touch. The approach not only reduces hardware complexity and cost, but also supports a more sustainable generation of AI-enabled products. The technology has since been adopted and evaluated by global companies spanning automotive, consumer electronics and robotics.

Giving machines a sense of touch

Making touch-responsive materials typically involves intricate networks of sensors and electrical components. They’re expensive to make and add to the mountains of e-waste we produce.

In 2015, Guo and his co-founder, industrial designer Ming Kong, envisioned a different approach: what if the entire surface could act as a sensor? Could you replace the hardware with sophisticated AI, coupled directly to the surface?

Aside from simplicity – reducing cost and e-waste – this would also mean that almost any material, of any shape, could become a smart surface, and enable much larger smart surfaces than with conventional approaches. For Guo and Kong, this was a challenge enthusiastically accepted, and they founded TG0 after graduating from Imperial College.

Six years of R&D culminated in a deceptively simple-sounding process. First, the team analyses the surface to see if touching it changes any measurable properties, for example, how it conducts electricity and its temperature. Its proprietary AI model extracts and analyses signals from the surface and translates this into a signal map – like a heatmap.

So far, the company has made this work for a wide range of different materials, including polymers, as well as wood, glass and metal.

After the following five years commercialising the technology, TG0 has struck up an array of partnerships, including in robotics, consumer products and premium automotive brands. With its partners, the company is co-designing and engineering new types of human-machine interfaces able to interpret different types of touch, including pressure, gesture, tapping, sliding, twisting and even shear force. 

Along with allowing robots to physically ‘feel’, the technology could unlock greater intelligence from everyday objects. Imagine a smart car seat that can recognise a driver from their posture, alerting and enacting a safe mode if they are tired; or a smart hospital bed that can help monitor patients through their pressure and movements, helping to prevent bed sores.

Small but mighty AI

On top of creating a unique new class of human-machine interfaces, unlike anything we’ve seen before, the company’s approach to the AI under the hood is also strikingly different.  While AI has made enormous progress in vision and language, touch remains one of the least developed senses in machines.

When Guo and Kong founded the company, the foundations were being laid for the current large language model (LLM) boom. After his PhD, Guo turned down some job offers from top companies: they were working on LLMs, and he knew that low-cost and power-efficient AI would better suit his ambitions.

“Nearly nobody saw the potential of low-resource AI in 2015,” says Guo. “Everybody [was] talking about big data, about cloud computing.” Rather than following the crowd and sending raw data from surfaces to the cloud to be analysed by massive, resource-hungry AI models, he instead envisioned lightweight AI built into the products.

“We had to squeeze AI inside the hardware, make AI as lightweight as possible [so it] can run on low-cost chips,” Guo says. “At that time, nobody believed we could make that happen.” Now, TG0 is reaping the rewards of its choice: Guo says that making smart sensing in everyday products with LLMs would be “impossible” due to the energy costs. 

Less power-hungry AI models are welcome from a global perspective too, as by 2030, a UN report has warned that global data centres powering AI will consume triple the combined annual electricity use of Pakistan, Bangladesh and Nigeria.

While the years of R&D were a struggle at times, Guo describes a sense of duty to bring the technology into the world, driven by his belief in affordable, low-resource and sustainable AI. “AI should be affordable, accessible – then everybody can enjoy the benefits,” says Guo. “I never had any thoughts of giving up because of that belief.”