Skip to content

ROCKET-1

Visual-temporal prompting for embodied decisions

Visit official site

What rocket-1 is

The model is described as a vision-language model for embodied decision making in open-world settings.

What you can do with it

  • Guide policy actions with object segmentation
  • Predict actions with a causal transformer
  • Split complex tasks into smaller steps

What to watch out for

  • Past and current observations feed the protocol
  • SAM-2 handles real-time object tracking
  • Minecraft tests are described

Pros & cons

✓ What we like

  • GPT-4o and Molmo break complex tasks into steps
  • A Minecraft benchmark evaluates zero-shot generalization

! What to watch out for

  • Experiments are described inside Minecraft
  • Setup asks for code, docs and a test environment

FAQ

What protocol is used?

Visual-temporal context prompting is named.

Which trackers are involved?

SAM-2 is listed for real-time object tracking.

Last reviewed: 2026-11-15

More LLM API platform tools

View all →

How we review