My love for robots started with movies. First I, Robot, then Big Hero 6, and later Andy Weir’s Project Hail Mary. I read the book before the film, and yes, that little guy up in the corner is Rocky.
At some point I stopped just watching robots on screen and started trying to build them. I got into robotics competitions pretty early and ended up winning 30+ of them at a national level. That was probably the moment I realised, okay, I’m going to spend my life around robots somehow.
I later studied at the
University of Bristol and spent a lot of time at the
Bristol Robotics Laboratory, one of the biggest robotics labs in the UK. While I was there, I worked on
intelligent bionic legs for people with lower-limb amputation, alongside Paralympic athletes who needed prosthetics that could actually keep up with them.
That changed how I thought about robotics. The best robots don’t always look like sci-fi. Sometimes they just help someone get a piece of their life back.
Since then, the thing I’ve cared about most is building robots and AI systems that are actually useful. I love the flashy demos too, they’re fun and they show what might be possible. But the harder question is: what happens after the demo? Can the system work reliably in the real world, around real people, with all the messiness that comes with that?
That question pulled me toward vision. If robots are going to act in the world, they first need to understand what’s happening around them. At Almetra (previously Deltia), I worked on a 7B vision-language model for spatio-temporal action segmentation. Basically, you show it a video and it breaks down what’s happening, action by action, over short and long time horizons.
We deployed this across 60+ factory sites for companies including ABB, helping teams see patterns on the factory floor that were previously invisible. For me, that felt like a step toward factories that can actually understand themselves.
Now I’m back on the robotics side, building systems that can act on that understanding. As Robotics Lead at Almetra, I work on the physical-intelligence layer for real robots, especially complex manipulation tasks that need to work outside clean lab demos.
The part that still feels magical to me is robot learning: watching a robot learn a behaviour instead of hand-coding every tiny step. I’ve worked with classical control, like 1 kHz impedance control, and with learned approaches like diffusion policies, vision-language-action models, and world models. My honest view is that the best systems will probably be somewhere in the middle. Robots need learning, but they also need structure, safety, feedback, and respect for physics.
I’m especially interested in world models: the idea that real intelligence comes from understanding how the physical world behaves, not just predicting the next token. But the truth is, nobody really knows the full recipe yet. We still don’t know how to deploy general-purpose robots at scale that do genuinely useful work.
That’s what makes the field so exciting.
Robots are where AI finally gets tested against reality. The question I keep coming back to is simple: what actually works at scale?
Right now I’m building toward that at Almetra, where I lead our collaborations with frontier robotics partners including Google DeepMind Robotics, MassRobotics, AWS, and NVIDIA. I also write about robotics and physical intelligence on my blog, mostly for people who are as obsessed with useful robots as I am.
