Ai2’s MolmoAct model ‘thinks in 3D’ to challenge Nvidia and Google in robotics AI
1 min read
Summary
Physical AI, the space where robotics and large language models meet, is seeing increased interest from large companies including Google, Meta and Nvidia, which are researching how robots can be augmented with LLMs .
To challenge these companies, the Allen Institute for AI (Ai2) has released an open source model called MolmoAct 7B, which allows robots to “reason in space”
While traditional vision-language-action (VLA) models do not think or reason in space, MolmoAct’s 3D spatial reasoning capabilities mean it can apply anywhere a machine needs to reason about its physical surroundings, according to Ai2.
The institute said its Benchmark testing showed the model had a task success rate of 72.1%, beating competitors including Google, Microsoft and Nvidia.