Kanaria Tech

[Kanaria Tech @kanariatech]

This week, @alisher_ai joined Akamai's invite-only leadership lunch for six companies, with @kanariatech as the only physical AI company in the room.

Akamai's leadership shared their vision for the company's future, with a strong focus on edge computing. That direction resonates with us. Our robots navigate busy, people-filled spaces with on-board intelligence, and the cloud adds extra capability on top. The closer that compute sits to our robots, the better.

Today, Kanaria Tech runs three workloads on Akamai Cloud's RTX PRO 6000 Blackwell GPUs:
🔁 Training residual RL policies for KRM
📊 Benchmarking each KRM version across environments and scenarios
🤖 Serving larger KRM model that complement our on-board KRM model

Thank you to the Akamai team, including their SVP of Products @JonAlexander, Cloud Computing CTO @JayJenkins, and Regional VP @SidharthPisharoti, for the invitation and the open conversations. Special thanks to @SidharthPisharoti for his interest in deepening our collaboration. We look forward to exploring more ways to grow together, especially across APAC.

Big thanks to the Japan team for continuous support @YusakuMatsuura, @HitoshiKaneko and @TakashiIto 🦾

More on our upgraded KRM architecture coming soon 👀

#PhysicalAI #Robotics #EdgeComputing #Akamai #KanariaTech

[Kanaria Tech @kanariatech]

We've filed an international patent application for KRM (Kanaria Robotic Model)🤖

Here’s the problem we’re solving:

A person steps into a robot’s path.
The robot sees an obstacle.
It stops. Waits. Stops again.

Multiply that across a busy hospital or warehouse, and those stops add up fast.
KRM works differently.

Instead of just reacting to where people are, it anticipates where they’re going, and moves smoothly around them.

What makes that possible:

📎 One model that understands the scene and decides the robot’s moves together
📎 It doesn’t just show what it sees. It explains, in plain language, what it’s thinking and why it’s moving the way it is: “Slowing down, person ahead on the left.”
📎 It learns mostly from everyday video, not expensive labeled robot data
📎 The same core model works across different types of robots

💡 That second point matters more than it might seem. Moving smoothly is only half of social navigation. The other half is people understanding what a robot is doing and why. That’s where trust between humans and robots begins.

The goal is simple: robots that move through our spaces the way a considerate person would.

Huge thanks to the team who made this happen. More to come 🚀

#Robotics #AI #EmbodiedAI #Innovation #Patents

[Kanaria Tech @kanariatech]

🎬 Kanaria Tech: Company Trip, aka The Reality Show

We laughed. We fought. We broke eggs. We almost broke bones. We got emotional.

👀 We saw each other from angles we never expected.

And somehow… we got closer ❤️

Full episode below👇
#KanariaTech #TeamTrip #StartupLife #CompanyCulture #TeamBonding

https://t.co/VkHeEMDNzp

[alphaXiv @askalphaxiv]

“World in World: Explore the World with World Models”

Video world models can generate long rollouts, but controlling them from new viewpoints usually needs task-specific training or adapters.

This paper instead turns source frames, geometry, and past generated states into visual evidence that a frozen world model can directly read through self-attention.

This then gives training-free camera-controlled rerendering with better long-horizon consistency, unseen-view completion, and lower camera error than prior methods.

https://t.co/BORSNbgoWV

[Kanaria Tech @kanariatech]

For most of history, our tools freed our bodies but this is the first one that thinks for us. This talk explores what it means to stay human in the age of artificial intelligence, where answers are instant and thinking can be outsourced.

Through simple but powerful examples, it challenges whether we are losing something deeper. Because the real question isn’t what AI can do, but what we choose to do with it.

Our CTO @alisher_ai shared his views. Watch the talk through the link below:
https://t.co/pu4BCaxmi4

[World Labs @theworldlabs]

Introducing Atlas:

The world's first multimodal world model that generates image and video frames with pixel-perfect camera control and reconstructs them in 3D.

Model the world, move the camera, and simulate space & time.

[Hirokatsu Kataoka | 片岡裕雄 @HirokatuKataoka]

"Visual General Intelligence: A White Paper"
is now on arXiv! https://t.co/IP8e477sdM

We've discussed:
- What is Visual Intelligence?
- Can we develop it toward General Intelligence (VGI)?

In the VGI white paper, we bring together diverse perspectives on Visual Intelligence and discuss how intelligence grounded in visual modalities and experiences could develop toward General Intelligence.

Beyond proposing a single definition of VGI, we introduce, propose, and discuss different perspectives on what Visual Intelligence could be and how we could develop it toward General Intelligence.

This paper is a starting point for a broader discussion on the future of Visual Intelligence.

Many thanks to all the great co-authors!

[Kanaria Tech @kanariatech]

What a day! Our Tokyo office open house is officially in the books 🎉

Huge thanks to the investors, business partners, family, and friends who came out to celebrate with us.

To everyone who has believed in Kanaria Tech's mission to build robots that understand people — thank you for your time, your questions, your support, and your energy. Days like this don't happen without you.

Here's to the next chapter🤖🧡

hashtag#Robotics hashtag#AI hashtag#Kanaria hashtag#OpenHouse hashtag#Tokyo hashtag#StartupJourney

[Kanaria Tech @kanariatech]

Ozgur Yilmaz, Assoc. Prof., Head of AI, just walked in with a box of Gaziantep baklava - and a very confident claim: Turkish baklava is the original, and everything else is just a tribute act😅

We're not here to start a baklava war. What we will say: nothing brings a team together faster than someone sharing a piece of home.

Grateful for teammates like Ozgur who bring both deep AI expertise and Gaziantep's finest pistachios to the table🇹🇷

#TeamCulture #KanariaTech #AIResearch #robotics

[Kanaria Tech @kanariatech]

Home turf. 🏠
GIC CONNÉCT celebrated the 1st anniversary of the Grassroots Innovation Center — home base for Kanaria Tech. Our CEO Keito Takishita took the stage at Ritsumeikan University's Biwako-Kusatsu Campus, sharing his story as both founder and current Ritsumeikan student.
None of this would be possible without Ritsumeikan University and Kusatsu City, who've supported us immensely — from Ritsumeikan's financial backing to Kusatsu's continued partnership, including close contact with the mayor's office.
Thank you to everyone who stopped by our booth. Kusatsu, you feel like home. 🚀
#GICConnect #Kusatsu #Ritsumeikan #Robotics #Shiga

[Alisher Abdulkhaev @alisher_ai]

Thank you, @ATPinsights @MichaelWaitze for having me today. Really enjoyed the discussion!

Building great Physical AI products takes serious engineering. We @kanariatech envision a socially aware robotic brain built on a dual-system approach.

Looking forward to a world with socially aware “robot friends” moving around us 🤖

[Michael Waitze @MichaelWaitze]

On today's episode of ATP Insights #46, we discussed how @Kimi_Moonshot reportedly has access to a cluster of roughly 20,000 @nvidia Hopper-generation chips through a computing agreement with @AlibabaGroup.

The infrastructure has become an important part of the company’s ability to train and operate its Kimi models, including Kimi K3.

libaba has denied supplying Moonshot with H200 chips specifically, but it has not disputed the broader computing relationship.

On Today's show:

00:00 - Intro
03:20 - Moonshot Has a 20,000 Nvidia Chip Cluster
31:40 - @ASvanevik, CEO and Co-Founder at @nansen_ai
43:24 - Alisher Abdulkhaev (@alisher_ai), Co-founder & CTO at @kanariatech
01:03:26 - Kurt Liu, CEO at @perxona_ai
01:24:35 - @sonam_pankaj_ , CEO at StarlightSearch
01:42:50 - Ethan Lee, Founder at Smalt AI
02:03:13 - Ben Styles, Chief Commercial Officer at @SuperAPIdev
02:23:39 - @Bhagwat_Pant, Consultant at CXGLOBE

https://t.co/CAxlOVyd5h

[Kanaria Tech @kanariatech]

How did we solve the data challenge in robotics?
Watch the full video on YouTube: https://t.co/RoXXNCbONC

[Kanaria Tech @kanariatech]

@ATPinsights @alisher_ai we are excited and looking forward to listening to your talk;)

[Kanaria Tech @kanariatech]

A trained brain isn't enough.
Our new video breaks down two features that make KRM-powered robots feel natural to be around:

🤖 Anticipation — the robot reads where people will move next, and plans ahead instead of reacting late.
💬 Explainability — the robot tells you what it's about to do, and why.
Together, they form a two-way interface: the robot reads people, people read the robot. That's what makes navigation feel smooth instead of jerky.

Watch here 👇 https://t.co/w7PwozzSf8

Next up: Scalability & Flexibility — the features that make KRM easier to commercialize across many robots.

@alisher_ai
#Robotics #AI #FoundationModel #KRM

[Kanaria Tech @kanariatech]

Thanks @dealroomco for featuring us among Japan’s hottest and promising AI companies 🥳

(Thanks for the referral @ikulyatin)