MolmoSpaces: A Large-Scale Open Ecosystem for Robot Navigation and Manipulation
This open-source ecosystem could finally solve robotics' biggest bottleneck.
A large research team has introduced MolmoSpaces, an open ecosystem for training and benchmarking robot policies at unprecedented scale. It features over 230,000 diverse indoor environments and 130,000 annotated objects, including 48,000 manipulable items with 42 million pre-computed stable grasps. The simulator-agnostic platform supports the full spectrum of embodied AI tasks, from navigation to long-horizon manipulation. Initial benchmarks show a remarkably strong 0.96 sim-to-real correlation, validating its utility for scalable robot learning.
Why It Matters
It provides the massive, diverse training data needed to build robots that can generalize to the real world's endless variations.