ToyBlocks Enables Controlled Robotic Eval: Depth vs Color, Compositionality, Planning
allen_ai · x · 2026-11-06
Allen AI describes the controlled experiments ToyBlocks supports: whether depth input improves robot control versus color images; whether models can stack four blocks without training on four-block stacks; and how models perform at planning long tasks. Researchers can also design their own evals.
More from Embodied
- RoboPapers to host author mixer at CoRL conference — chris_j_paxton · 2026-11-06
- New Robot Learning Tutorial Video with Oxford PhD Francesco Capuano — yacinelearning · 2026-11-06
- Lambda and Fei-Fei Li's lab present I3L: +68.3-point success in mobile manipulation — TheZachMueller · 2026-11-06
- Embodied AI's real question isn't System-1 vs System-2, but drawing capability boundaries — ZhidingYu · 2026-11-06
- Farewell Cybertruck: how its 48V and by-wire tech enables the Cybercab — mitchdeg · 2026-11-06
- Autonomous shipbuilding robots place, fit and weld hull parts in one click, 5:35 per operation — GabGarrett · 2026-11-06