From Beijing to “exponential”: multi-robot dashboard, faster onboarding, and better 3D

Beijing recap, then Europe next

Nils just returned to Hong Kong after his first Beijing trip since COVID (he lived there 2009–2017). Four reasons for the trip:

1. Lecture at Guanghua MBA School (Peking University)
2. Meetings with Galbot and Robotera (“robot shopping”)
3. An Intercognitive meetup in Beijing with surprisingly high-caliber turnout
4. Preparing for a Europe trip to try to close a long-running deal (“fingers crossed”)

A big robotics company is now sending us leads

A standout moment this week: one of the largest robotics companies “slid into his DMs” with a serious lead. “There is this very big account in Asia. They want what you’re cooking… and they’re looking at robots. Let’s do this together.”

That’s the flywheel we want: robots are hard to deploy, and what we bring (fast spatial setup + shared maps + navigation) helps them sell more robots.

Also this weekend in Hong Kong:

• Saturday: a robot lab visiting for a follow-up
• Sunday: @Robo_Tuo, the “robot sherpa of Shenzhen” visiting the lab

Internal demo day highlights

1) Multi-robot orchestration (“Legion” → future Cactus robotics dashboard)

Phil demoed an early multi-robot orchestration tool on top of the real world web:

• Shared “allowed navigation” areas (purple)
• Two robots shown live (blue + green), both Padbot-family but from different bases/vendors
• A “promo round” task kicked off for one robot while the other continues patrolling

Why it matters:

• Robot makers generally don’t offer true multi-robot coordination across heterogeneous fleets.
• This becomes a customer-facing robotics dashboard in 2026: battery, health, mode (patrol, return-to-base), and soon fault alerts.

The punchline: once setup drops from weeks to minutes/hours, customers can justify multiple robots per store (especially in big footprints) and actually manage them sanely.

2) Retail promo robots become ROI-realistic

We talked through a concrete near-term use case: promo robots (mobile retail media displays paid for by brands).

• Retailers and brands ask for this often
• Historically too painful to deploy for ROI
• With our flow: quick DMT mapping (even without full reconstruction) + robot sign-in in under ~2 minutes makes it finally viable.

3) “Eight hours vs three months”: onboarding is becoming memetically fit

A recurring theme: competitors can take months to set up a store. We’re seeing customers do meaningful installs without us on site.

They referenced a robotics company that needs ~5 minutes per SKU to set up. At 30,000 SKUs, that’s 2,500 hours of work — completely impractical.

In contrast:

• A customer’s reconstruction aligned extremely well with their existing floor plan
• The store's domain was created in ~8 hours (same day), self-serve, no Auki staff
• Another example: a customer arrived mid-afternoon, sent for reconstruction by ~8pm, and had a full reconstruction a few hours later — again, entirely self-run (their own infra / nodes, no Auki systems)

They called it out explicitly: the system is becoming “memetically fit” — easy enough to spread inside real orgs via training and repetition.

3D pipeline progress: Gaussian splats + subtractive scanning

Gaussian splatting (photorealistic “robot memory”)

Robin shared Gaussian splats generated from the same capture flow as DMT. The important bit is that reconstruction becomes a foundation:

• Capture like normal DMT
• Use reconstruction as initialization
• Produce a more photorealistic scene representation
• This becomes another node type / pipeline layer over time

He framed it as: reconstruction is one step, and the output can feed many other CV processes.

Subtractive scanning (maps that can forget)

JB showed “subtractive scanning,” a major long-term requirement for collaborative maps:

• Observations carry confidence scores
• If repeated rescans no longer support a feature (e.g., a removed marker/QR), its confidence drops until it disappears from the map

This is core to maintaining living environments without manual cleanup.

Dimensional + G1 demo momentum

A spicy update dropped mid-call: Dimensional thinks they can have an early joint demo by end of next week: A whole-body G1 walking around and pointing at items (Terri-style).

A small “standards” moment that mattered

At an Intercognitive meetup, Nils got pushback from a Chinese VC on global standards. The response that flipped the tone:

• We’re aiming to deploy ~500 robots in 2026
• That implies ~$20M in hardware spend
• When you’re spending that much with robot vendors, they listen — and you become the standard by deploying at scale.

Wrap

The vibe of the week: this is about to go exponential because deployment friction is collapsing (hours not months), fleet coordination is starting to exist, and the 3D pipeline is maturing into something maintainable over time.

アウキ・ラボについて

Aukiはポーズメッシュという地球上、そしてその先の1000億の人々、デバイス、AIのための分散型機械認識ネットワークを構築しています。ポーズメッシュは、機械やAIが物理的世界を理解するために使用可能な、外部的かつ協調的な空間感覚です。

私たちの使命は、人々の相互認知能力、つまり私たちが互いに、そしてAIとともに考え、経験し、問題を解決する能力を向上させることです。人間の能力を拡大させる最も良い方法は、他者と協力することです。私たちは、意識を拡張するテクノロジーを構築し、コミュニケーションの摩擦を減らし、心の橋渡しをします。

ポーズメッシュについて

ポーズメッシュは、分散型で、ブロックチェーンベースの空間コンピューティングネットワークを動かすオープンソースのプロトコルです。

ポーズメッシュは、空間コンピューティングが協調的でプライバシーを保護する未来をもたらすよう設計されています。いかなる組織の監視能力も制限し、空間のプライベートな地図の自己所有権を奨励します。

分散化はまた、特に低レイテンシが重要な共同ARセッションにおいて、競争優位性を有します。ポスメッシュは分散化運動の次のステップであり、成長するテック大手のパワーに対抗するものです。

アウキ・ラボはポスメッシュにより、ポーズメッシュのソフトウェア・インフラの開発を託されました。