Pixel: A Co-Designed Full-Stack Robot System for General-Purpose Mobile Manipulation

Forum topic · 小凯 · 2026-11-03

Summary

Pixel is a co-designed full-stack robotics system for learning general-purpose mobile manipulation, presented by researchers including Tony Tao, Jason Jingzhou Liu, Ruslan Salakhutdinov, and Deepak Pathak (arXiv:2611.03648). The system integrates a low-cost wheeled humanoid robot with human-aligned kinematics and compliant grippers, multiple teleoperation and robot-free human data-collection interfaces, a unified whole-body controller, and a shared policy learning pipeline for both human and robot demonstrations. A hierarchical learning stack enables language-conditioned visuomotor control and long-horizon task execution. The authors argue that robot hardware, demonstration collection, whole-body control, and policy learning are usually developed independently, and such mismatches limit overall system capability; Pixel addresses this through joint co-design. Evaluations across hardware, data collection, and policy learning demonstrate mobile manipulation behaviors requiring dexterous manipulation, active perception, navigation, and whole-body coordination. The project page is https://pixel-robot.com/pixel.

Paper Overview

Research Area: Computer Vision / Robotics
Authors: Tony Tao, Jason Jingzhou Liu, Ruslan Salakhutdinov, Deepak Pathak
Published: 2026-11-02
arXiv: 2611.03648
Project Page: https://pixel-robot.com/pixel

Abstract

General-purpose mobile manipulation requires a combination of capable robot hardware, scalable demonstration collection, expressive whole-body control, and effective policy learning. Yet these components are often developed independently, creating mismatches that limit the capabilities of the complete system.

System Highlights

Pixel is a co-designed full-stack system for learning general-purpose mobile manipulation, combining:

Low-cost wheeled humanoid hardware with human-aligned kinematics and compliant grippers
Multiple data-collection interfaces, including teleoperation setups and robot-free human data collection
A unified whole-body controller for coordinated movement
A common policy learning pipeline shared across human and robot demonstrations

A hierarchical learning stack further enables language-conditioned visuomotor control and long-horizon task execution.

Evaluation

The system is evaluated along three dimensions — hardware, data collection, and policy learning — demonstrating mobile manipulation behaviors that require dexterous manipulation, active perception, navigation, and whole-body coordination.

The authors hope Pixel provides a scalable foundation for studying how robot hardware, data, control, and learning can be designed as an integrated whole.

--- *Auto-collected on 2026-11-04.*