Robotics Lead at Almetra, London UK β building Almetra's robotics division from the ground up, turning factory video intelligence into robot learning systems for industrial manipulation.
It started with a 7B VLM watching real work across 60+ factory sites using custom action-recognition models. The natural next step: closing the loop, so the same understanding that observes the work also drives the arms doing it.
The stack β VLAs for task decomposition, diffusion policies for contact-rich control, world models for predictive planning and sim-to-real, and classical impedance control at 1 kHz where the physics is already solved. Object-centric, cross-embodiment by design. Working on both end-to-end and hybrid approaches to see what actually ships.
I lead technical collaborations across frontier robotics labs including Google DeepMind, NVIDIA, AWS, and MassRobotics.
- π€ Dexterous manipulation policies from demonstrations, robot interaction data, and multimodal task context
- π Diffusion policies for contact-rich control; world models for predictive planning and sim-to-real
- ποΈ 1 kHz impedance control where classical methods still win
- π Fine-grained video understanding of manufacturing processes
