WRC2026 Observations: The Pivotal Moment Robotics Shifts from Performance to Practicality

From Dancing to Working: The Wind Has Shifted at Robot Expos

I just got back from WRC2026 (World Robot Conference), and as an ordinary attendee, my most immediate impression was this: this year's show was noticeably more pragmatic.

At past editions, humanoid robots mostly danced and waved on stage — symbolic gestures with limited practical value. This time around, more and more robots were attempting real work: tidying rooms, brewing coffee, retail picking, warehouse operations, and long-horizon manipulation tasks. Most of these capabilities are powered by VLA (Vision-Language-Action models) or World Models, with software increasingly driving hardware evolution.

Interestingly, many robots on the floor stumbled and made mistakes — and that was actually reassuring. It strongly suggested these weren't teleoperation demos, but genuine autonomous runs. More importantly, some robots demonstrated the ability to understand their current context and adapt their strategies accordingly. That's what real intelligence looks like.

Data Collection: The New Battleground for Embodied AI

If last year's keyword was "hardware," this year's show introduced another unavoidable theme: "data."

Physical-world data collection is heating up dramatically. UMI (Universal Manipulation Interface), ego-centric video, and teleoperation-based data collection were nearly ubiquitous across vendor booths. The underlying logic is straightforward: if robots are ever going to achieve genuine generalization, real-world data is becoming just as critical as the hardware itself — perhaps more so.

This mirrors the trajectory of large language models. Once model architectures converge, data quality and scale become the decisive competitive moat. Embodied AI is hitting a similar inflection point.

A Significant Gap Remains Before Reliable Mass Production

Despite the clear progress, most practical demonstrations still felt like "functional" rather than "reliably deployable at scale."

Take dexterous hands as an example: a single dexterous hand's BOM (Bill of Materials) can exceed 1,500 components. Manufacturing consistency is a massive challenge — any deviation along the supply chain can compromise overall reliability and yield. This is the chasm that embodied AI must cross before it can transition from lab to industry.

That said, compared to robots that could only "dance on stage," actually attempting to complete real tasks represents a genuine leap forward.

Emerging Embodied AI Players Worth Watching

I spent most of my time at WRC2026 on the smaller, lesser-known vendors and startups. A few left a strong impression:

SynapX

One of the rare vendors to demo an EMG (electromyography) wristband live on-site, alongside a World Model-driven robotic arm capable of tasks like folding clothes. EMG technology holds enormous potential for human-robot interaction and data collection.

Galaxea (银河通用)

Visitors could place orders on the spot, with the robot autonomously identifying and grasping products — and generalizing across different SKUs without needing per-category adaptation. This kind of generalization is the key to real-world deployment in retail and warehouse settings.

Morphi Robot

Uses a hybrid World Model + VLA architecture capable of sustaining complete long-horizon tasks lasting up to 15 minutes. The ability to reliably complete long-horizon tasks is a critical benchmark for measuring a robot's true intelligence.

AgileX Robotics (松灵机器人)

Multiple model companies on the floor were using their robotic arms. AgileX also showcased a "one-to-many" teleoperation + ego-centric data collection solution — addressing a genuine pain point in today's data acquisition pipelines.

Dexforce

A World Model + humanoid robot combination capable of tasks like brewing coffee, demonstrating the possibilities of general-purpose manipulation.

Looking at these players' technical approaches, "World Model + VLA" is clearly emerging as the mainstream architectural consensus in embodied AI.

A Fragmented Industry: When Will Standardization Arrive?

However, one issue impossible to ignore: the industry remains highly fragmented.

Every company is building its own hardware and defining its own data formats. This fragmented approach enables rapid experimentation in the short term, but in the long run it leads to massive duplicated effort and wasted resources. The question naturally arises: when will embodied AI see genuine standardization and technological convergence?

Looking at the histories of the PC and smartphone industries, standardization has typically been a prerequisite for large-scale industrial breakout. Whoever first establishes widely accepted standards — in data formats, interface protocols, or hardware modules — could well dominate the next phase of the industry.

Seeing Robots Fall Actually Builds Confidence

Of course, there were plenty of "fail moments" on the floor — humanoid robots that fell and couldn't get back up, with staff scrambling to diagnose the problem.

But frankly, I was glad to see those debugging sessions. Compared to perfectly choreographed, flawless performances, real errors and live troubleshooting are precisely what shows these robots are running and iterating in the real world. That kind of "imperfection" actually gave me more confidence, not less.

Embodied AI is at a pivotal inflection point — shifting from "demonstration" to "practical utility." It's not perfect yet. Challenges around manufacturing consistency, reliability, and industry standardization remain unsolved. But the direction is becoming increasingly clear.