Ground-Up VLA

The Ground-Up VLA Substack is my attempt to publicly build an understanding of vision-language-action (VLA) models from first principles. I have little background in machine learning, so the series starts with the underlying systems and machine-learning concepts before moving toward robot foundation models.

I took a break from this to work on another side project in the late spring of 2026: Second Silicon.

The posts below are listed in reading order, from oldest to newest.

• Ground-Up VLA: Introduction
• Ground-Up VLA: GPU Basics
• Ground-Up VLA: Neural Networks
• Ground-Up VLA: Self-Attention
• Ground-Up VLA: Deep learning and transformers
• Ground-Up VLA: LayerNorm and Residual Addition
• Ground-Up VLA: Feed-Forward Sublayer
• Ground-Up VLA: Embedding and Positional Encoding
• Ground-Up VLA: Turning Model Outputs into Answers
• Ground-Up VLA: Training Concepts
• Ground-Up VLA: Simple Modeling with TorchLite and PyTorch
• Ground-Up VLA: Model Depth and Width
• Ground-Up VLA: Batch Size and Regularization
• Ground-Up VLA: Multi-Head Attention
• Ground-Up VLA: Optimizers and Learning Rate Scheduling
• Ground-Up VLA: Encoders, Decoders, and Cross-Attention