Posts
Other
About

Ted Staley

I am a senior AI engineer at Johns Hopkins University Applied Physics Laboratory (APL), where I work in AI and robotics. My work has spanned many areas, primarily reinforcement learning, model pretraining, and imitation learning. I am interested in understanding approaches to robotics that are general and scalable, and leverage data sources that have low barriers to collection. Some of my projects result in publications, which you can find on my Google Scholar.

In addition to my RL and robotics work, I spent several years working in LLMs. I designed the data processing and pretraining routines for APL's first in-house LLMs: multibillion parameter models trained on trillions of tokens. Concurrently I taught ChatGPT from Scratch at the Johns Hopkins Engineering for Professionals (EP) Program, a graduate course offering a deep breakdown of building LLMs in PyTorch.

I stood up this website to collect my thoughts and notes on AI and robotics, and to track my publicly released projects and publications. I am also maintaining repositories of my relevant code implementations on my github.

You can reach me at ewmstaley@gmail.com

All Recent Posts:

An 8x8 Touch Sensor

Reconstructing FlexiTac, Part Two

May 25, 2026

Thoughts on Hand Capture

And Reconstructing FlexiTac, Part One

May 23, 2026

On Shuffling Tokens

Preparing Trillion-Token Datasets

May 12, 2026

LLMs, MatSci, NeurIPS 2025

Coupling GPT with Materials Synthesis Simulation

March 12, 2026

GAIL with Pixels Only

Rewarding for Visual Fidelity

May 16, 2025

GAIL

Rewarding for Fidelity

April 29, 2025

MuJoCo Cronenbergs

(Mis)Adventures in Style Transfer, Part 2

February 10, 2025

MuJoCo CycleGAN

(Mis)Adventures in Style Transfer, Part 1

January 27, 2025

Flowing with Fewer Steps

Shortcut Models Notes and Review

December 12, 2024

Going with the Flow

Notes on Flow Matching (Policies)

December 9, 2024

Modeling the World

RSSM & TSSM Notes and Experiments

December 1, 2024

Diffusion Policy Part 3

Playing CarRacing-v3 with Diffusion

November 1, 2024

Older Posts: