Twenty-three selected works spanning metalevel reasoning, representation and abstraction, ADAPT, predictive vision, and robot mental/world models. Each work has its own portable citation page.
Continues the predictive world-model lineage using Unreal Engine 5 to maintain a robot mental model synchronized with the physical environment.
Uses a synchronized 3-D virtual world as a common representation for physical understanding, cognitive understanding, prediction, and human–robot collaboration.
Extends predictive internal world modeling to human–robot coordination, using active perception, simulation, and task-dependent visual context to support interaction.
Maintains a real-time virtual copy of robot and environment, compares expected and observed data through Match-Mediated Difference, and uses simulation of possible futures for planning.
Compares real video with synthetic imagery generated from an internal simulation so behaviorally significant differences can be detected and tracked efficiently.
Connects Soar’s cognitive mechanisms with Robot Schemas for concurrent real-time sensorimotor control, providing a bridge between symbolic cognition and embodied robotics.
Uses a dynamic 3-D internal world model to predict what a robot should see; discrepancies between prediction and observation drive active visual processing and obstacle avoidance.
Introduces ADAPT (Adaptive Dynamics and Active Perception for Thought), a cognitive architecture designed specifically for robotics and organized around concurrent sensorimotor activity and active perception.
Identifies concurrency and active perception as core requirements for embodied cognition and describes ADAPT’s integration of Soar, Robot Schemas, and representation reformulation.
Foundational work on task-preserving abstraction: changing a planning representation while preserving the structure that matters to the task, with the goal of reducing planning and sensing cost.
Reports progress on a goal-directed cognitive vision system that assembles local spatial observations into a dynamic world model.
Studies how a mobile robot can classify observed human behaviors and use internal models to predict subsequent actions.
Investigates how an explicit 3-D representation of the environment affects robot-vision processing.
Treats robot vision as goal-directed search: saccadic and vergence movements build local 3-D models that are integrated into a larger representation of the environment.
Uses cognitive representations and simulation to classify observed behaviors, connecting perception to a robot's interpretation of actions.
Investigates how cognitive semantics can connect perceptual representations to action in behavior-based robotic systems.
Describes the human–robot interaction components of ADAPT, including the integration of natural language with models of human behavior.
Presents a predictive robot-vision approach in which an internal model generates expectations and perception focuses on differences between predicted and observed scenes.
Examines the metalevel as a system level and the kinds of knowledge and representations needed for a system to reason about and improve its own reasoning.
Develops task-representation decomposition as a way to reorganize autonomous problem solving and expose structure useful for abstraction and planning.
Extends ADAPT around predictive mental models, active perception, symbolic reasoning, and foundation models used as perceptual primitives under cognitive control.
Examines how intelligent global behavior can emerge from simple local actions when the internal organization of actions reflects essential structure in the environment.
Develops an algebraic approach to identifying and formulating reusable structural patterns in problem solving, extending the representation and reformulation research line.
Deductive representation reformulation using semigroup theory; publicly archived by NASA.