GPT-6 Astra builds interactive simulation from a single image, wowing researchers

akapoor_av8r · x · 2026-09-13

• Quoting @Parvkpr's hands-on test: GPT-6's Astra can generate an interactive simulation environment from a single image zero-shot, with surprisingly faithful results — even the liquid's color was nearly exact (though it couldn't simulate the reaction).
• The poster adds that pairing it with GRID works even better (GRID + Astra > Astra).
• Context: real2sim research (single image/video → interactive simulation) is heating up for evaluation and training use cases, e.g. NVIDIA's SimFoundry (code still unreleased after months of waiting). Abilities that once required entire research papers are now one prompt away.

Original post →

More from Embodied

• UCLA's ChairNav teaches autonomous wheelchairs to learn each rider's preferences — zhoubolei · 2026-10-31
• Amazon Echo kills local voice storage: everything goes to the cloud from March 28 — alex_verem · 2026-10-31
• HP paid $116M for Humane, then bricked every $699 AI Pin 10 days later — alex_verem · 2026-10-31
• Moxie, the $800 AI robot friend for kids, went silent overnight after shutdown — alex_verem · 2026-10-31
• Subvocal's YC-backed glasses decode silent speech at 200+ wpm, built in 4 months — garrytan · 2026-10-31
• Nomagic warehouse robots go from bimanual to trimanual, pause to solve tricky picks — m_wulfmeier · 2026-10-31

Embodied channel →