Post-Training Intern
Noematrix
Post-trained a UMI-pretrained world-action model for real-robot pick-and-place and deployed it on a dual-arm robot.
- Role
- Post-Training Intern
- Dates
- Aug 2026 – Sep 2026
At Noematrix, an embodied-intelligence company building vision-language-action models, I post-trained FastWAM, a world-action model pre-trained on UMI data, for real-robot pick-and-place tasks such as placing eggs into trays and returning tools to toolboxes.
I deployed the post-trained checkpoint to a dual-arm real robot through the team's inference server, integrating it with the gripper and device pipeline. Real-robot pick-and-place success rose from 15% to 78.3%. The evaluation work here is what motivated robostats.
I return to Noematrix over winter break 2026 to work on agentic robotics.