Robotics · Cognitive Architectures · Planning

Predictive cognition for embodied intelligence

Research connecting task representation, cognitive architectures, mental and world models, active perception, hierarchical planning.

Browse publicationsExplore research lineagesPaceRobotLab GitHub

Research programs

Research program

Cognitive Architectures & ADAPT

ADAPT integrates symbolic cognition, Robot Schemas, concurrent sensorimotor processes, active perception, learning, and predictive internal representations for embodied intelligence.

Research program

Mental Models, World Models & Active Perception

A long-running research line in which robots maintain predictive internal models, generate expected observations, direct perception toward informative discrepancies, and use simulation to plan.

Research program

Representation, Abstraction & Hierarchical Planning

Research on task-preserving abstraction, representation reformulation, metalevel reasoning, and learned hierarchical structure for more efficient planning.

Older work on metalevel representation and task-preserving abstraction connects to current hierarchical planning; ADAPT’s active perception and dynamic virtual worlds connect to contemporary predictive mental/world models and foundation-model-enabled embodied intelligence.

Research Overview

2004 — ADAPT establishes a robotics-specific cognitive architecture.
2006 — Predictive vision uses a dynamic 3-D world model to generate expectations.
2009–2012 — Real and synthetic imagery are compared; Match-Mediated Difference supports model correction and planning.
2015–2018 — 3-D world models support human–robot coordination and shared spatial understanding.
2026 — Unreal Engine 5 work continues the robot mental-model lineage; ADAPT 2.0 extends the architecture toward foundation models.