---
title: "Discussion on \"π0: A General-Purpose Robot Policy via VLM + Flow Matching — Physical Intelligence's First Answer\" | Hashnode"
description: "Discussion on \"π0: A General-Purpose Robot Policy via VLM + Flow Matching — Physical Intelligence's First Answer\". TL;DR\nπ0 (pi-zero) is a general-purpose robot policy model released by Physical Intelligence in October 2024. The core idea: combine a pre-trained VLM (PaliGemma 3B) with a Flow Matching-based continuous action output — inheriting Internet-scale sema..."
url: "https://hashnode.com/posts/p0-a-general-purpose-robot-policy-via-vlm-flow-matching-physical-intelligences-first-answer/69e099d83b61c7ddf86a7ebf"
---

[Feed](https://hashnode.com/feed) /

Discussion

# [π0: A General-Purpose Robot Policy via VLM + Flow Matching — Physical Intelligence's First Answer](https://telos-robotics.hashnode.dev/p0-a-general-purpose-robot-policy-via-vlm-flow-matching-physical-intelligences-first-answer?utm_source=hashnode&utm_medium=feed)

TL;DR π0 (pi-zero) is a general-purpose robot policy model released by Physical Intelligence in October 2024. The core idea: combine a pre-trained VLM (PaliGemma 3B) with a Flow Matching-based continuous action output — inheriting Internet-scale sema...

Apr 16 · 9 min read · telos-robotics.hashnode.dev  
Read the article

-   [#flow-matching](https://hashnode.com/tag/flow-matching)
-   [#machine-learning](https://hashnode.com/tag/machine-learning)
-   [#robot-learning](https://hashnode.com/tag/robot-learning)
-   [#robotics](https://hashnode.com/tag/robotics)
-   [#vla](https://hashnode.com/tag/vla)

## Written by

[telos](https://hashnode.com/@telos)

@telos · Seoul, South Korea

  
[View profile](https://hashnode.com/@telos)

I build robots for a living. Not in simulation. Not in a lab. On the floor, with real hardware, real failure modes, and real deadlines. My work spans the full stack of modern robotics: embedded systems firmware, autonomous navigation, and the end-to-end pipeline for Vision-Language-Action (VLA) model development — dataset collection, training, inference optimization, and sim-to-real transfer. I've worked hands-on with platforms from Unitree, Deep Robotics, Dexmate, and Robotis, among others. Each one teaches you something different about the gap between what models promise and what robots actually do.

## Responses

Markdown supported · ⌘↵ to post

No responses yet.