Frontier Tech

Sequenced learning

Paths across the frontier.

A path orders difficult ideas by dependency, then turns them toward a concrete systems outcome. Available and planned work remain clearly distinct.

Active learning paths

1 available4 planned

Embodied Agents

A systems path from language-conditioned planning through skill accumulation, grounding, control, and safe deployment.

Outcome

Design and critique an embodied-agent stack with explicit interfaces for planning, memory, feedback, verification, control, and safety.

View the Embodied Agents path
  1. 01

    Language models as planners

    Planned
  2. 02

    Voyager: compounding capability

    Available
  3. 03

    Multimodal grounding and world models

    Planned
  4. 04

    Hierarchical robot control

    Planned
  5. 05

    Safety for open-ended agents

    Planned