JEPA World Models: From Pixels to Physics

Series

JEPA World Models: From Pixels to Physics

A deep-dive series documenting the development and stress-testing of JEPA architectures in low-cost robotics.

[3] Post-Mortem Analysis: Why My First World Model (JEPA) Is "Blind"
The pipeline is ready, the data is converted, and the GPU has completed its first "stress test." Now comes the most critical phase for any research engineer: the autopsy of the latent space. In this t
Apr 15, 202611 min read10
[2] Data Engineering: Why and How to Convert LeRobot (Parquet/MP4) to HDF5
In my previous post, I explained why the JEPA architecture is such a promising lead for robotics. But between Yann LeCun’s theory and the first \(loss.backward()\), there is a massive wall: the data.
Apr 15, 20264 min read11
[1] Rethinking Robotics: Why I’m Betting on JEPA over VLAs
To test my research intuitions, I’m documenting my work on a JEPA (Joint-Embedding Predictive Architecture) world model, starting with low-cost robotics datasets like LeRobot and the Koch SO-ARM101. T
Apr 15, 20265 min read33