Ted Troxell Get in touch

Physical AI

Models that have to obey physics.

I work at the forefront of physical AI: machine learning that senses, predicts and acts in the real world. For thirteen years I've built it and led the teams that ship it, from defense hardware and welding cells to robots and factory tooling. Everything here runs live in your browser. Anything orange, you can move.

Open to speaking, ML leadership roles and select advisory work.Get in touch

Scroll to begin.

01 · Sense

Every model starts with a measurement.

Throw the pendulum. Each frame the page records its angle and how fast it's turning, the way a sensor on a real machine would.

Angle θ
0°
Angular velocity ω
0
Frames recorded
0
Length
1.00 m

02 · Learn · Measure

Acceleration, plotted against angle.

Each dot is one frame. The model never sees the pendulum's equation, only this cloud of measurements. The shape is already a clue.

02 · Learn · Build a library

Eight guesses at what matters.

Every measurement becomes eight candidate terms the answer might use: a constant, θ, ω, sin θ, cos θ and some products. One row per frame, one column per term, with the acceleration to predict on the right.

02 · Learn · Fit everything

Least squares weighs every term.

With every term allowed, the weight smears across lookalikes. The fit is close, but the equation is a mess no one could reason about.

α = …

02 · Learn · Prune

Physics is usually sparse.

Drop the terms under the threshold, refit, repeat. What survives is the equation, and the dashed pendulum now runs on it.

α = …

Candidate terms

03 · Act · Manipulation

An arm that solves its own reach.

A four-axis arm picks up the block and sets it in the tray as you scroll. Every frame it solves inverse kinematics: the joint angles that put the gripper exactly where it needs to be.

Drag the orange block before the arm grabs it.

03 · Act · Mapping

A rover that maps as it drives.

A laser sweeps the room. Space the beam passes through becomes free; the point it hits becomes wall. Here the rover knows its own position; full SLAM estimates that too.

Drag the orange crate into its path.

03 · Act · Aerial control

A drone that learns the wind.

A feedback controller holds the drone on its figure eight, and an online estimator learns the wind from the drone's own tracking errors, then cancels it.

Drag the orange fan to blow it off course.

04 · Make

Where the plastic goes.

Scroll to inject. Melt enters this 150 × 90 mm plate at the orange gate and fills by the shortest path through the cavity. Dotted lines are weld lines, where fronts meet: the weak spots a good gate location moves.

Drag the orange gate. Shortest-path fill is a first-order stand-in for a flow solver.

05 · Lead

2013 – 2017

Engineer & Product Manager · Think-A-Move

Designed, built and managed novel AI technology for the US Department of Defense.

Physical world

05 · Lead

2017 – 2018

Data Scientist · Lincoln Electric

Built systems that automatically describe and log welding events on the assembly floor.

Physical world

05 · Lead

2018 – 2020

CEO · liftr

Ran the company as CEO, based in Greater Cleveland.

Leadership

05 · Lead

2020 – 2021

Chief Technology Officer · LiftCamp, Inc.

Led technology and engineering as CTO.

Leadership

05 · Lead

2021

Director of Data Science · Captiv8

Directed the data science function.

Leadership

05 · Lead

2021 – now

Member of Technical Staff · Atomic Industries

Machine learning for manufacturing, in Cleveland.

Physical world

05 · Lead

The region

Northeast Ohio

Most of my career has been built here. Thirteen years, from defense hardware to the factory floor, and through CLE AI & DATA, a network of nearly 5,000 people across the region.

06 · Give back

CLE AI & DATA

Northeast Ohio's premier AI and data community: nearly 5,000 members, grown from a local meetup into a hub of the Midwest AI Corridor, with one mission, to make AI accessible to everyone. As its President and CEO, I bring my enterprise network to the table. Sponsors keep the annual event free for the whole community.

Raised
$100,000+
Members
~5,000
Partners, past and present
31
Annual event
Hundreds attend, free

Drag the orange hub. Partners and audience shares are from the CLE AI & DATA sponsorship prospectus.

CLE AI & DATA · What it leaves behind

Impact that outlasts the event

More than $100,000 raised, put to work so Northeast Ohio's AI community stays open to everyone.

  • Free
    An annual event anyone can attend

    Hundreds of people every year at the Cleveland Museum of Natural History, most recently on 28 September 2026. Sponsorship keeps it free.

  • Growth
    From a local meetup to a Midwest hub

    Nearly 5,000 members, and a hub anchoring the broader Midwest AI Corridor.

  • Access
    Learners in the same room as decision makers

    The audience is 58% developers and technical leads, 25% decision makers and 16% learners and re-skillers, all in one place.

  • Funded
    Events, workshops and connections

    More than $100,000 raised and put to work on events, hands-on workshops and the connections between the people building AI in the region.

Past and present partners

  • AWS
  • OpenAI
  • Google
  • IBM
  • Cisco
  • Palantir
  • Snowflake
  • MongoDB
  • Elastic
  • Cloudera
  • Hortonworks
  • Nutanix
  • Arista
  • Wiz
  • Kaggle
  • The Linux Foundation
  • KeyBank
  • Progressive
  • Expedient
  • CGI
  • Centric
  • Intricity
  • Further
  • Velera
  • EOX Vantage
  • Moreland Connect
  • AI Rising
  • Atomic Industries
  • Greater Cleveland Partnership
  • Case Western Reserve University
  • Cleveland State University

Partner with CLE AI & DATA

2026 sponsors: AWS, OpenAI, Nutanix, Arista Networks, Further, CrossComm, EOX Vantage, SkillSpout, Cisco and Velera. cle-ai.com

Speaking

Talks with live physics on stage

The demos on this page come with me. I speak about physical AI, and about building and leading the teams that put it into production.

  • Topic
    Teaching neural networks physics

    What physics-informed models get right, where they break, and when a classical solver still wins.

  • Topic
    From the lab to the factory floor

    Lessons from shipping machine learning into welding cells, defense hardware and tooling.

  • Topic
    Founder to function lead

    What running a startup taught me about leading ML teams inside larger companies.

Invite me to speak

For organisers

Ted Troxell works on physical AI as a Member of Technical Staff at Atomic Industries in Cleveland, and is President and CEO of CLE AI & DATA, a Northeast Ohio community of nearly 5,000 members. He has built AI for the US Department of Defense and welding analytics at Lincoln Electric, and has served as a CEO, a CTO and a director of data science.

Download headshot (JPG)

Writing

Notes and projects

Occasional posts on models, simulation and what breaks when the two meet. These are carried over from the current site.

  • 5 Nov 2023
    What’s a PINN?

    A Physics Informed Neural Network (PINN) is a specialized type of artificial intelligence model that brings together the power of traditional neural networks with…

  • 27 Aug 2023
    Similarities of Graphs and Images in Computer Vision with Graph Neural Networks

    In the vast realm of computer vision, where every pixel and relationship carries meaning, the amalgamation of graphs, images, and Graph Neural Networks (GNNs)…

  • 27 Aug 2023
    What are Graph Neural Networks?

    In a world interconnected by a web of relationships, understanding and harnessing the complexities of these connections is becoming increasingly important. Enter Graph Neural…

  • 9 Jul 2023
    The Significance of Physics-Informed Neural Networks

    In recent years, the marriage of physics and machine learning has given rise to a revolutionary concept known as Physics-Informed Neural Networks (PINNs). These…

Ted Troxell

About

Hi, I'm Ted.

I've built machine-learning and statistical models since 2013, starting with AI for the US Department of Defense and welding analytics at Lincoln Electric. I've also been a CEO, a CTO and a director of data science. The work I like most is where a model has to respect the physical world: conservation laws, materials, noisy sensors.

Today I'm a Member of Technical Staff at Atomic Industries and President and CEO of CLE AI & DATA. Based in Cleveland, I'm open to speaking invitations, ML leadership roles and select advisory work.

Start a conversation

Elsewhere

Step 1 of 2

What can I help with?