Trossen Robotics

We asked our team one simple question
“What got you excited about robots?”

We’re pausing our usual tech focused content to share something a little more human.
Meet the people behind the robots at Trossen Robotics and hear what pulled them into this work.

We’ve published a new technical walkthrough showing how to train and deploy Pi-0 and Pi-0.5 ML models using the OpenPi framework, now integrated with Trossen AI arms.

This video covers the full workflow, including data collection, model training and configuration, key parameters for fine-tuning, and running trained policies. We also discuss practical considerations such as batch size selection, adapter-based fine-tuning, inference settings, and the current limitations when working with small or task-specific datasets.

OpenPi provides a flexible foundation for imitation learning and robot policy training, and this walkthrough is designed to help researchers and developers move from data collection to deployment more efficiently.

As machine learning tools rapidly evolve, robotics research needs data pipelines that keep pace. This explains how shifting from custom systems to widely adopted frameworks has dramatically improved the way researchers work. Moving from ROS based data collection to newer platforms like Hugging Face and the Bot framework allowed tighter integration with action chunking, transformers, diffusion policies, and the tools researchers already rely on.

By building our hardware directly into these pipelines, teams can train models faster, create benchmarks with less effort, and avoid rebuilding entire ML stacks. The result is more time spent on innovation and far less on integration.

Full Episode
https://youtu.be/rdxyEO1Ugpo