Embodied intelligence truly entering industry isn't about whether a demo can dance — it's whether robots can run continuously for ten hours at a customer site without errors. At the 2026 World Robot Conference (August 19-23), Galaxea used the largest booth on the floor (500 sqm) to give its answer: no parkour, no backflips — every demo revolved around "real work."
Nexo: A 30-DoF Embodied Platform
Nexo is Galaxea's newly released flagship wheeled-arm humanoid robot. Its full body has 30 degrees of freedom covering dual arms, shoulder/neck, waist, and chassis, with joints coordinated in real time under force-position hybrid control. Dual-arm peak payload is 20 kg (10 kg per arm), with 8 hours of battery life. It supports fast charging, battery swapping, and autonomous recharging — enabling 7x24 continuous operation.
But the 30 DoF isn't the point; what it does is: one machine适配 four scenarios — e-commerce retail, industrial manufacturing, express logistics, and commercial services. In other words, Nexo is not a "specialized device" but a "general-purpose productivity platform" — a route that diverges from Unitree, Zhiyuan A2, and UBTECH Walker.
World's First Robot-Operated Front Warehouse Taking Online Orders
At WRC, Galaxea built a 1:1 real commercial front warehouse. Users placed random orders via a mini-program, and the robot autonomously recognized items, planned paths, grasped and picked, and bagged and packed — completing the full closed loop from order to delivery. This is claimed to be the "world's first robot front warehouse supporting online ordering with fully automated closed-loop fulfillment."
The hard part isn't grasping but "generalization + flexibility + multi-stage":
- Ten-thousand-scale SKUs require no per-item programming; the model recognizes and perceives autonomously.
- The robot schedules multi-stage continuous tasks on its own, self-correcting and recovering when facing environmental changes, task interruptions, or execution anomalies.
- Fine manipulation of deformable objects — handling the variability of plastic bags — breaks through the flexible-operation bottleneck in real logistics fulfillment.
- 1 (top-level core): Dual-wheel drive of "whole machines + intelligence." Galaxea Dynamics accumulates hardware capabilities (bodies, modules, motion control, systems engineering); Galaxea Model carries the embodied foundation model and task intelligence.
- 3 (three product lines): Kengo (general bipedal humanoid, for unstructured complex scenes), Nexo (general wheeled-arm robot, for structured scenes), Lemo (dual-arm desktop manipulation platform, for the developer ecosystem).
- N (thousand-industries deployment plan): Partner-driven rollout via the Xingtu Plan and Excellence/Pioneer Partner programs, with cooperation agreements already signed with 32 industry-leading companies.
- G0.5 foundation model embeds a "native chain-of-action-thought," unifying image perception, language understanding, logical reasoning, and action generation in a single autoregressive data stream. It comprehensively surpasses industry SOTA across 8 international benchmarks including LIBERO and RoboTwin 2.0. G0.5 MAX is coming soon, and the G0.5 reproduction plan has launched — model weights, inference APIs, benchmarks, and full fine-tuning tooling are fully open.
- Fast-WAM world model skips the "generate future video" step, cutting single-step inference latency from ~800 ms in the traditional paradigm to 190 ms — a 4x+ speedup. A pretrained large model based on the Fast-WAM architecture will also be released.
- G-Fleet distributed reinforcement learning system supports cluster-level policy push, parallel real-robot rollouts, real-scene data backflow, human-intervention calibration, distributed RL training, and automated evaluation and version release — closing the loop of "model deployment → real-world interaction → data curation → iterative training → new version release," the key to safe, efficient continuous evolution after large-scale real-robot deployment.
According to on-site data, the system has completed field testing at front warehouses of multiple leading Chinese enterprises, processing over 2,000 items of sorting and handling per day.
Another World First: Robots Assembling Robots
Elsewhere in the exhibit, Galaxea showed another "world's first": a robot independently completing a high-precision, long-horizon assembly task of "centimeter-scale screw insertion + automated drill screwdriving." The robot identifies and grasps centimeter-scale screws among complex parts, precisely inserts them into narrow holes, then switches to an automated drill, aligning with the hole center, controlling posture, applying stable pressure, and completing the fastening. The process continuously invokes visual positioning, dual-arm coordination, force-controlled contact, and motion planning.
Compared with traditional manual teaching, operation speed improved by over 30%. The real meaning of that number: the robot is no longer merely "able to do one fine manipulation once" — it has the hard capability for fine-grained, long-horizon, high-stability mass production in real scenarios.
The "1+3+N" Next-Generation Productivity Brand Strategy
Founder and CEO Gao Jiyang officially announced the "1+3+N" strategy in a keynote:
Three Core Technology Assets
The Numbers
Galaxea has delivered thousand-unit-scale orders for model-driven real productivity, with a delivery target of 10,000+ units next year. In a pilot at an automotive parts manufacturer, a robot fleet running Galaxea's system raised production line efficiency by 40% and cut labor costs by 65%.
COO Li Tianwei sketched the product evolution path: primarily whole-machine sales before 2026; from 2026 H2 to 2028, a solution-subscription model where customers pay for industry solutions for specific production scenarios, with the robot as merely the physical carrier of the solution. In other words, robots are shifting from "product" to "service."
"Being able to demonstrate" is only the starting point; "really working" requires composure in the face of the unexpected. In the second half of embodied intelligence, what gets tested is not performance ability, but the capacity to work reliably.
References: China News Service report (Aug 23), Little Bear Finance "Galaxea Shines at WRC 2026" (Aug 22), Stone Jin on-site interview at IPO Early Know, Galaxea "From Models to Productivity" technology and industry forum, Gao Jiyang keynote via Tech Beat, public LIBERO/RoboTwin 2.0 evaluations.