The Year in AI Papers
The year/Independent research

Paper 2602.07837

RLinf-USER: A Unified and Extensible System for Real-World Online Policy Learning in Embodied AI

Hongzhi Zang, Shu'ang Yu, Hao Lin, Tianxing Zhou, Zefang Huang, Zhen Guo, Xin Xu, Jiakai Zhou, Yuze Sheng, Shizhe Zhang, Feng Gao, Wenhao Tang, et al.

Published
Feb 2026
Research lab
Independent
Citations
4
GitHub
Not linked

01 In brief

Summary

USER is a unified and extensible system for real-world online policy learning in embodied AI, addressing the challenges of physical-world training such as lack of acceleration, reset, and replication.

It treats robots as first-class hardware resources alongside GPUs via a unified hardware abstraction layer, enabling automatic discovery, management, and scheduling of heterogeneous robots.

An adaptive communication plane uses tunneling-based networking, distributed data channels, and SM-aware weight synchronization to handle cloud-edge communication.

The learning framework is fully asynchronous with a persistent, cache-aware buffer for long-horizon experiments and crash recovery.

It supports extensible abstractions for rewards, algorithms, and policies, including CNN/MLP, flow-based generative policies, and large VLA models.

Experiments on five real-world manipulation tasks (Peg Insertion, Charger, Cap Tightening, Pick-and-Place, Table Clean-up) show high performance, with RLPD achieving near-perfect success within 2000s, and HG-DAgger improving π0 success from 39/60 to 58/60 on Pick-and-Place.

The system enables multi-robot coordination, heterogeneous manipulators, edge-cloud collaboration, and long-running asynchronous training, offering a systems foundation for real-world policy learning.

02 From the paper

Abstract

Online policy learning directly in the physical world is a promising yet challenging direction for embodied intelligence. Unlike simulation, real-world systems cannot be arbitrarily accelerated, cheaply reset, or massively replicated, which makes scalable data collection, heterogeneous deployment, and long-horizon effective training difficult. These challenges suggest that real-world policy learning is not only an algorithmic issue but fundamentally a systems problem. We present USER, a Unified and extensible SystEm for Real-world online policy learning. USER treats physical robots as first-class hardware resources alongside GPUs through a unified hardware abstraction layer, enabling automatic discovery, management, and scheduling of heterogeneous robots. To address cloud-edge communication, USER introduces an adaptive communication plane with tunneling-based networking, distributed data channels for traffic localization, and streaming-multiprocessor-aware weight synchronization to regulate GPU-side overhead. On top of this infrastructure, USER organizes learning as a fully asynchronous framework with a persistent, cache-aware buffer, enabling efficient long-horizon experiments with robust crash recovery and reuse of historical data. In addition, USER provides extensible abstractions for rewards, algorithms, and policies, supporting online imitation or reinforcement learning of CNN/MLP, generative policies, and large vision-language-action (VLA) models within a unified pipeline. Results in both simulation and the real world show that USER enables multi-robot coordination, heterogeneous manipulators, edge-cloud collaboration with large models, and long-running asynchronous training, offering a unified and extensible systems foundation for real-world online policy learning.