Tanguy Pauwels (@tanguy-pauwels)

[3] Post-Mortem Analysis: Why My First World Model (JEPA) Is "Blind"

Explore why a JEPA world model fails with low data. A technical autopsy of latent space, SIGReg constraints, and scaling strategies for robotics.

11 min readblog.tanguy-pauwels.fr

Apr 15

[2] Data Engineering: Why and How to Convert LeRobot (Parquet/MP4) to HDF5

Why MP4 is bad for JEPA training and how to build a memory-efficient HDF5 conversion pipeline for LeWM World Models.

4 min readblog.tanguy-pauwels.fr

Apr 15

[1] Rethinking Robotics: Why I’m Betting on JEPA over VLAs

Explore why JEPA could outperform VLA models in robotics. A deep dive into latent dynamics, physical hallucinations and the path to System 2 reasoning

5 min readblog.tanguy-pauwels.fr

Apr 15

Tanguy Pauwels

About

Available for

Tanguy Pauwels's blogs

[3] Post-Mortem Analysis: Why My First World Model (JEPA) Is "Blind"

[2] Data Engineering: Why and How to Convert LeRobot (Parquet/MP4) to HDF5

[1] Rethinking Robotics: Why I’m Betting on JEPA over VLAs

Tanguy Pauwels

About

Available for

Tanguy Pauwels's blogs

[3] Post-Mortem Analysis: Why My First World Model (JEPA) Is "Blind"

[2] Data Engineering: Why and How to Convert LeRobot (Parquet/MP4) to HDF5

[1] Rethinking Robotics: Why I’m Betting on JEPA over VLAs

Tanguy Pauwels

About

Available for

Tanguy Pauwels's blogs

[3] Post-Mortem Analysis: Why My First World Model (JEPA) Is "Blind"

[2] Data Engineering: Why and How to Convert LeRobot (Parquet/MP4) to HDF5

[1] Rethinking Robotics: Why I’m Betting on JEPA over VLAs

Search Hashnode

Tanguy Pauwels

About

Available for

Tanguy Pauwels's blogs

[3] Post-Mortem Analysis: Why My First World Model (JEPA) Is "Blind"

[2] Data Engineering: Why and How to Convert LeRobot (Parquet/MP4) to HDF5

[1] Rethinking Robotics: Why I’m Betting on JEPA over VLAs