Research

Exploring the frontiers of artificial general intelligence

Research Areas

World Models & Representation Learning

We develop architectures capable of learning abstract representations of the physical world. Unlike traditional generative models, our approaches compress essential information while ignoring non-predictive variations.

Latent Representations State-Space Models Self-Supervised Learning Discrete World Models

Agent Planning & Reasoning

Our agentic systems use world models to simulate the consequences of their actions before executing them. This enables sophisticated planning and causal reasoning in complex environments.

Model-Based RL Hierarchical Planning Causal Reasoning Goal-Conditioned Policies

Memory & Continual Learning

We build persistent memory architectures allowing systems to maintain extended context and accumulate knowledge over time, without catastrophic forgetting.

Memory Networks Continual Learning Episodic Memory Knowledge Retention

AI Safety & Control

Safety is at the core of our research. We develop formal guardrails, human-in-the-loop mechanisms, and behavior guarantees for critical applications.

Safe Exploration Constrained Optimization Human-in-the-Loop Formal Verification

Multi-Modal Perception

Our models integrate data from multiple sensory modalities, enabling rich and nuanced understanding of the physical world in all its complexity.

Vision & Language Sensor Fusion Cross-Modal Learning Embodied Perception