Research

Pioneering Efficient AI

Our research focuses on developing foundational models that enable embodied AI systems to perceive, reason, and act in the physical world while consuming minimal energy. We publish our findings to advance the field and contribute to sustainable AI development.

Research Areas

Our work spans multiple interconnected areas, all aimed at creating AI systems that are both capable and sustainable.

Sparse Neural Architectures

Developing neural network designs that activate only the parameters needed for each input, dramatically reducing energy consumption while maintaining performance.

5 publications

Physics-Informed Learning

Incorporating physical laws and constraints into model training to reduce data requirements and improve generalization in embodied AI systems.

3 publications

Efficient Multimodal Fusion

Creating unified representations that combine visual, tactile, and proprioceptive inputs for robots with minimal computational overhead.

4 publications

Real-Time Decision Making

Enabling sub-millisecond inference for responsive robotic control through optimized model architectures and deployment strategies.

2 publications

Publications

Selected publications from our research team. We believe in open science and share our findings with the broader AI community.

Energy-Efficient Transformers for Embodied AI: A Sparse Attention Approach

E. Chen, L. White, M. Taylor
Working Paper (2025)

We present a novel sparse attention mechanism designed to significantly reduce energy consumption while maintaining competitive performance on embodied AI benchmarks.

Physics-Constrained Neural Networks for Robot Manipulation

M. Taylor, E. Chen
Preprint (2025)

A new approach to training manipulation models that incorporates physical constraints, aiming to reduce training data requirements substantially.

Efficient Multimodal Representations for Autonomous Systems

E. Chen, L. White
Working Paper (2025)

We introduce a unified multimodal architecture designed to process visual, tactile, and proprioceptive inputs with low-latency inference.

Sustainable AI: A Framework for Energy-Aware Model Development

L. White, E. Chen, M. Taylor
Technical Report (2024)

A comprehensive framework for developing AI systems that optimise for both performance and energy efficiency.

Interested in Collaboration?

We actively collaborate with academic institutions and research labs around the world. If you're interested in partnering with us on research projects, we'd love to hear from you.

Get in Touch