05 Hierarchical RL & Skill Learning

How can complex behaviour emerge from reusable skills?

Area 05: Hierarchical RL & Skill Learning

Master Question

How can complex behaviour emerge from reusable skills?

What We Want to Discover

Long horizon tasks are hard to learn one primitive action at a time. This area studies how an agent can discover, represent, and reuse temporally extended skills, and how a higher level controller should choose among those skills to solve tasks that were never encountered as a whole during training.

Why It Matters

Reusable skills are what let learning transfer across tasks instead of starting from zero every time. Without a principled account of hierarchy, complex behaviour has to be hand engineered task by task, which does not scale.

Core Concepts

  • Options framework
  • Subgoal discovery
  • Skill composition and skill chaining
  • Feudal and manager worker architectures
  • Temporal abstraction

Relevant Disciplines

  • Cognitive science, particularly motor skill acquisition
  • Linguistics, particularly compositional structure
  • Control theory

Potential Mimicry Sources

  • Motor skill hierarchies in animals and humans
  • Modularity in biological motor control
  • Assembly of complex behaviour from simple reflexes in insects

Projects

No projects yet.

Active Questions

No projects yet, so there are no active derived questions to report.

Key Findings Across Projects

Pending. No projects in this area have produced findings yet.

Unresolved Questions

Pending.

Connections to Other Areas

Pending. Connections will be identified as projects in this area develop.