Announcement
Ditto-Bench tests GPT-6 Astra and MolmoAct2 on tougher robot-control tasks
Its creators say the benchmark pairs simple goals with challenging physics to test both models beyond simple robot-control tasks.
TLDR
Ditto-Bench’s creators say GPT-6 Astra has shown impressive results on simple robot-control tasks. They built the benchmark to test Astra and MolmoAct2 on harder physics, including geometric constraints, precise contact and improvised tool use. They also examined design choices that may help the agents succeed or leave them stuck, and explored possibilities such as in-context learning for robotics.
Combined views
—
5 Sources, first seen ago
