New labs, research teams, model families and technical infrastructure — resolved to canonical entities, linked to primary sources, and paired with a reason to spend attention.
An omni world model pretrained for text, images, video and 3D using a multimodal autoregressive diffusion transformer.
It unifies controlled generation, sparse-view reconstruction and real-to-sim robotics inside one spatial context.
A new Beijing lab for system-level demonstration, validation and joint development across robotics and physical AI.
It connects embodied-model progress to chips, sensing, control and deployable robotics systems.
A university–industry lab focused on next-generation agentic AI for Korean language and domestic industries.
It pairs academic research with frontier compute and a clear regional deployment agenda.
A £30m UK lab spanning Oxford, UCL and Imperial to develop open, human-centred, resource-efficient AI for the real world.
Its remit explicitly goes beyond LLMs into robotics, engineering, healthcare and scientific discovery.
A frontier research group dedicated to large-scale embodied world-model pretraining for autonomous humanoids.
Its proposed data flywheel spans web video, egocentric footage, simulation, teleoperation and on-policy robot data.
A new HAI center building an empirical science of how AI changes jobs, teams, coordination and organizational performance.
It treats workplace AI as a measurable organizational system rather than a collection of productivity anecdotes.
A dedicated team for moving research previews such as Claude Code, MCP, Skills and Cowork into new product categories.
It is a useful organizational signal: agent infrastructure and human-computer interaction are becoming a first-class lab agenda.