Research

Research

Our research follows the layers of silicon life — the chips it thinks with, the nervous system it senses with, the body it acts through, and the communities it learns in.

01

Domain-Specific Accelerators

Hardware that lets silicon life think fast on little energy — FPGA and edge accelerators, novel low-precision number formats and quantization for large language models, and AI-assisted chip design tools.

  • FPGA
  • Low-precision arithmetic
  • LLM quantization
  • AI for EDA
02

Edge Perception & Interaction

Sensing and decision-making that run on the device itself — multimodal perception, speech and vision, and natural interaction with the people around it, in real time and without the cloud.

  • Multimodal perception
  • On-device inference
  • Speech interaction
  • HCI
03

Robot Bodies & Robot Learning

Robot platforms that act safely in everyday places — mechanical and embedded design, motion and gesture, and learning methods that let robots acquire new skills from demonstration and experience.

  • Robot platforms
  • Embedded control
  • Robot learning
  • Manipulation
04

Real-World Data & Collective Learning

Data is the experience of silicon life, and sharing it is the layer we are starting on: egocentric data capture with our EGO devices, datasets for robot learning, and, next, methods for robots to learn from one another. We also grow an open community of makers around these tools.

  • Egocentric data
  • Robot learning datasets
  • Open hardware

Beyond these layers, we explore human–AI co-creation — see Lyra.

See Publications and Projects.