// research portfolio · james thomas

The physics of intelligent behavior

I am an MS Artificial Intelligence candidate at UT Austin where I have studied the full breadth of relevant AI topics, but have also gone deep in the most exciting areas of frontier research: Simulated physics, LLMs, energy-based models, embodied agents, as well as the less celebrated areas of validation, benchmarking, and validation.

This site is a place where I can publish the research that I choose to go even deeper on.

MS AI · UT Austin Energy-based models Sim-to-Real
// research

What I'm exploring

Most of my projects are the same question in different clothes: how does energy and dynamics govern the way a system evolves and acts over time..

Physical simulation
Energy-based models
Embodied control
Language & reasoning
Physics Informed AI
the central force · hover to pause
// work

Selected work

001

A* vs MDP

Pathfinding under uncertainty — compare A* search against a Markov Decision Process on the same grid.

pathfindingexplore
View ↗
002

Simulation environments

Physics-accurate environments for stress-testing control policies before they meet real hardware.

embodied controlin progress
View ↗
// writing

Recent research log

A* vs MDP: Pathfinding Under Uncertainty ↗ jamesthomas.dev — full writing archive