Researchers suspect Astra was trained on 3D and robot action data, giving it stronger 3D awareness
georgiagkioxari · x · 2026-09-18
Robotics researcher Georgia Gkioxari argues Astra is more 3D-aware than other frontier multimodal models.
- She notes such capabilities typically come from pretraining or RLVR, so 3D data was likely in the training mix
- Michael Black adds that while Astra isn't "natively" 3D/4D, it understands enough to help scale 3D/4D data collection
- Gkioxari also notes robot action data was part of training
A first-hand observation from senior robotics researchers about Astra's likely training data composition.
More from Embodied
- Nano-texture iPhone Duo display makes Apple Pencil feel like writing on paper — rudrank · 2026-11-04
- NEURA launches Neuron AI, but Europe's Physical AI test is whether the stack matures together — ingliguori · 2026-11-04
- Baidu's Apollo Go wraps two-week public robotaxi test rides in Tung Chung, Hong Kong — Baidu_Inc · 2026-11-04
- imec neural probe integrates photonics directly into 130nm CMOS backend — jwt0625 · 2026-11-04
- Chris Paxton: interactive robot sim is the future but glaringly data-limited — Scobleizer · 2026-11-04
- DreamX: video diffusion models as universal latent action learners for robots — kaist-ai · 2026-11-04