Vision-language-action models hit a memory wall, and three labs are trying to break it

The latest VLA models are starting to fail in a specific, predictable way: they remember the last few seconds, and not much more. Researchers are racing to fix it.

Priya Nair · 23 May · 3 min

The world model versus the policy: a debate that is finally getting resolved

A long-running theoretical disagreement inside robotics research is starting to resolve in favour of one side. The implications are bigger than they sound.

Lena Park · 23 May · 3 min

Nvidia's Isaac GR00T moves from research demo to developer platform

Nvidia's humanoid robotics foundation model has been talked about for two years. The new SDK release is the first time it looks like a serious platform play.

Nadia Rahman · 23 May · 3 min

Open X-Embodiment was supposed to be a research curiosity. A year on, it is the default dataset for serious robot manipulation research.

Isaac Mendez · 23 May · 3 min

A new model from DeepMind achieves real-world manipulation performance using zero real demonstrations. The sim-to-real gap may be closing.

Lena Park · 4 May · 3 min

A growing body of research suggests that internet video is a better training signal for robot policies than expensive human demonstrations.

Priya Nair · 3 May · 3 min

An open platform for sharing robot policy checkpoints has attracted 200 uploads in its first week.

Nadia Rahman · 2 May · 3 min

New hardware, better data pipelines, and architectural improvements have collapsed the cost curve for robot model training.

Lena Park · 1 May · 3 min

A decade of work on simulated locomotion training is paying off. The same techniques are being adapted for hands.

Isaac Mendez · 30 Apr · 3 min

A Llama variant fine-tuned for embodied reasoning and robot control has landed. The open-source implications are significant.

Nadia Rahman · 29 Apr · 3 min