After several months in stealth, I’m publicly announcing Positronic Robotics.

I believe that AI Robotics – specifically end-to-end ML policies controlling robots – is going to be a huge deal. AI paves the path to truly universal machines: the same commodity hardware solving multiple tasks.

The Reality Check

Despite all the hype, this field is much earlier than most think. We are now where self-driving was in 2015 – real potential, but years of groundwork ahead. This is the truth that very few in the industry are willing to admit publicly.

Our Goal

Our goal at Positronic is to make professional-grade ML robotics approachable.

Over the last year, we trained ML policies, spoke with dozens of teams, and learned the painful truths. Teams still lack the basics: * Clean data pipelines * Scaling dataset management * Objective and reliable evaluation * Recipes for model finetuning * Handling inevitable distribution drift in production * Managing various hardware

This is what we are fixing.

What We Are Shipping

We are shipping a few pieces I’m proud of:

  1. Positronic Runtime (pimm/): A Python-native, immediate-mode runtime to wire sensors, controllers, inference, and GUIs without the pain of managing ROS launch files or other DSLs.
  2. Positronic Dataset (positronic.dataset): A pragmatic dataset layer for multi-rate episodes, lazy transforms, and a web viewer for triage.
  3. A Clean End-to-End Loop: Collect → Inspect/Curate → Train → Validate.

Get Involved

Check the repository and docs, kick the tires, and tell us what breaks.

We’re also working on something big and exciting for release in January 2026.