Research Areas
Openbird Research investigates six interconnected research areas spanning cognitive architecture, memory consolidation, sensorimotor learning, and social development.
Developmental Intelligence
Rather than initializing models with comprehensive adult-level encyclopedic knowledge, this area investigates whether an artificial learner can acquire grounded understanding through an unfolding developmental trajectory analogous to infancy and early childhood.
Central Open Questions
- Can complex cognitive capacities reliably emerge from simple starting priors without task-specific engineering?
- What developmental scaffolds accelerate sensorimotor and concept acquisition in artificial learners?
- How does an artificial individual maintain stability while continuously incorporating novel structures?
Learning & Adaptation
Explores non-stationary learning environments where an agent must accumulate knowledge over years of simulated or physical experience without erasing foundational sensorimotor skills.
Central Open Questions
- How can modular network architectures isolate new episodic adaptations from core procedural memory?
- What self-supervised objective functions promote robust developmental exploration without reward collapse?
Memory & World Models
Focuses on memory not as a static database lookup, but as an active generative reconstruction process that predicts environmental affordances and supports mental simulation.
Central Open Questions
- How do episodic traces crystallize into semantic generalizations over sleep-like consolidation phases?
- What representations allow an artificial agent to construct an autobiographical record of its own experience history?
Curiosity & Internal Simulation
Investigating internal drives that motivate an agent to explore when external rewards are absent, alongside offline replay routines that simulate unobserved possibilities.
Central Open Questions
- What information-gain metrics prevent artificial learners from becoming trapped in noisy or trivial entropy sources?
- Can counterfactual simulation during offline quiescence accelerate physical problem-solving skills?
Social Development & Language
Investigating how language and communication emerge as functional tools for shared agency and social coordination between human caregivers and artificial learners.
Central Open Questions
- Does communicative intent develop more effectively from collaborative physical tasks than passive text modeling?
- How does joint attention facilitate mutual grounding in novel sensory environments?
Embodied Intelligence
Grounding developmental cognition in physical constraints—gravity, friction, inertia, and sensory latency—initially in high-fidelity physics simulations with future robotic embodiment.
Central Open Questions
- Which perceptual representations transfer reliably from simulated developmental environments to physical robotic platforms?
- How do motor limitations shape cognitive concept formation in early development?