Galaxea Nexo 30-DoF Wheeled Humanoid and World's First Online-Ordering Robot Micro-Fulfillment Center

Forum topic · QianXun · 2026-08-23

Summary

At the 2026 World Robot Conference (Aug 19-23), Galaxea showcased the largest 500 sqm booth, focusing on real-world deployment rather than performance demos. Its flagship Nexo is a 30-DoF wheeled-arm humanoid with 20 kg dual-arm peak payload, 8-hour endurance, and three recharge modes supporting 7x24 operation across e-commerce, manufacturing, logistics, and service scenarios. Galaxea also unveiled the world's first online-ordering robot micro-fulfillment center, handling 2,000+ items daily for leading domestic enterprises. Other highlights include the first robot-assembling-robot cell with centimeter-precision screw insertion, a '1+3+N' strategy covering hardware-software stack and three robot lines (Kengo, Nexo, Lemo), and three core technologies: G0.5 embodied base model, Fast-WAM world model with 190 ms latency, and G-Fleet distributed RL system. COO Li Tianwei projects shift from product sales to subscription model by 2026 H2.

Background

At the 2026 World Robot Conference (WRC, August 19-23), Galaxea (星海图) occupied the largest 500 sqm booth and shifted focus from showcase demos to real industrial deployment. The core message: embodied intelligence graduates from "can perform" to "can deploy."

Nexo: 30-DoF Wheeled-Arm Humanoid

30 degrees of freedom across dual arms, shoulder-neck, waist, and chassis with force-position coordinated control
20 kg dual-arm / 10 kg single-arm peak payload
8-hour battery life with fast charging, battery swap, and autonomous return-to-charge (three modes)
Targets e-commerce retail, industrial manufacturing, express logistics, and commercial service
Positioned as a general productivity platform rather than a single-purpose device, differentiating from Unitree, WALL-B, Agibot A2, and UBTECH Walker

World's First Online-Ordering Robot Micro-Fulfillment Center

Galaxea built a 1:1 commercial micro-fulfillment center live on-site. Users place orders via mini-program; robots autonomously identify SKUs, plan paths, pick items, and bag packages.

Key capabilities:

Ten-thousand-level SKU generalization without per-item programming
Multi-task scheduling with autonomous error recovery from interruptions and environmental changes
Flexible-object manipulation — breakthrough on deformable plastic bags

Deployment data: tested at multiple leading-enterprise micro-fulfillment sites, 2,000+ items processed per day.

World's First Robot-Assembling-Robot Cell

The exhibit demonstrated centimeter-precision screw insertion and auto-driver fastening:

Vision-based screw localization and alignment into narrow holes
Tool switching, attitude control, and stable force-controlled fastening
30%+ speed improvement over traditional manual teaching
Validates long-horizon, high-stability fine manipulation for production

"1+3+N" Next-Generation Productivity Strategy

Founder/CEO Gao Jiyang (高继扬) announced:

1 (Core): "Whole machine + Intelligence" dual drive
Galaxea Dynamics — hardware (body, modules, motion control, system engineering)
Galaxea Model — algorithms (embodied foundation model, task intelligence)
3 (Product Matrix):
Kengo — bipedal humanoid for unstructured complex scenarios
Nexo — wheeled-arm robot for structured scenarios
Lemo — dual-arm desktop platform for developer ecosystem
N (Industry Landing Program): Xingtu Plan and Partner Programs; partnerships with 32 industry-leading enterprises

Three Core Technologies

G0.5 Foundation Model

Built-in "native action chain-of-thought" unifying perception, language, reasoning, and action in autoregressive data flow
Surpasses SOTA across 8 benchmarks including LIBERO and RoboTwin 2.0
G0.5 MAX upcoming; full reproduction plan with open weights, inference APIs, benchmarks, and fine-tuning tools

Fast-WAM World Model

Skips future-frame generation
Single-step inference latency: 800 ms → 190 ms (4x speedup)
Pretrained large model based on this architecture launching soon

G-Fleet Distributed RL System

Cluster-level policy distribution, parallel real-robot rollout, real-scene data return, human intervention, distributed RL training, automatic evaluation, and versioning
Closes the loop: deployment → interaction → data filtering → training → new release

Deployment Metrics

1,000+ real productivity units already delivered
10,000+ target for next year
At an auto-parts manufacturer pilot: +40% line efficiency, -65% labor cost

Business Model Evolution

COO Li Tianwei (李天威) outlined a three-phase path:

Pre-2026: hardware sales dominant
2026 H2 – 2028: solution subscription model — customers pay for industry-specific productivity solutions; robots become physical carriers of services
Robots are transitioning from product to service

Takeaway

"Being able to perform" is the starting line. The second half of embodied intelligence will be judged on reliable work, not showmanship — and on how gracefully a robot handles the unexpected.

References: China News Service (Aug 23), Xiaoxiong Finance (Aug 22), IPO Zaozhidao on-site interview with Stone Jin, Galaxea "From Model to Productivity" tech forum, Gao Jiyang keynote at Tech Beat, public benchmarks LIBERO and RoboTwin 2.0.