Ralf Gulde retweeted
Online shopping returns are one of the hardest warehouse tasks to automate. Here is why. 👟
Sereact’s AI-powered dual-arm robots can handle returned shoes, clothing and mixed items, even when packaging is messy or products arrive in unfamiliar conditions.
Powered by Cortex, the system combines visual understanding with robotic manipulation to identify, grasp and sort items without requiring individual training for every product.
The technology is already operating at Zalando returns facilities in Germany and Poland, with CEVA Logistics.
🏢 Companies: @SereactAI , @Zalando & @cevalogistics
🤖 Robot hardware: @Universal_Robot
🎥 Media: Raphael Levy — demo at @NVIDIAGTC Berlin 2026
⚠️ This content is shared for informational purposes only. CTO Robotics Media is a media platform and does not own or develop the technology shown. Credit belongs to the original creators.
#Sereact #Zalando #CEVALogistics #UniversalRobots #WarehouseAutomation #PhysicalAI #Robotics #CTORoboticsMedia
A year ago, automating returns seemed out of reach for robotics.
Today, it’s underway. With Otto Group, we’re bringing 10 Sereact dual-arm robots to their logistics center in Poland, with 15 more to follow.
Grateful to the @OttoGroup_Com and @nvidia for their trust and partnership, and incredibly proud of the entire Sereact team for making it happen.
Read more: sereact.ai/posts/otto-group-…
No two returns look the same. That’s exactly why we built Cortex.
Dual-arm picking, live at @Zalando x @cevalogistics
Ralf Gulde retweeted
What a week at @NVIDIAGTC.
New partnerships with @Teradyneinc, @Universal_Robot, @nebiusai, @nvidia, @cevalogistics, and @Zalando. Cortex live on the floor handling shoebox returns, and a spot in @JensenHuang’s keynote.
All while shipping more robots than ever.
#GTCBerlin
Ralf Gulde retweeted
Shared from GTC with the @creandum team what’s new at @SereactAI but it deserves more eyeballs.
Company had 2 phenomenal quarters back to back and now fully upgraded from VLAM to world action models.
Khiem is a legend. Went from intern to AI research lead and serves @TuscherMarc & the gang for +4 years.
Only worry: my sleep over office mattress setup after the board meeting this week got a little too fancy over the years…
When we started @SereactAI , people told us returns were impossible to automate.
This is Cortex handling @Zalando
returns at a @cevalogistics fulfillment center. It’s live and in production, every shift.
Proud of the team that made this real.
Ralf Gulde retweeted
Real customers. Live orders. Production at scale.
We’re teaming up with @Teradyneinc and @Universal_Robot to run Cortex on UR cobots. Hundreds of robots shipped across Europe, and now heading to North America.
See it live at the @nvidia booth, Hall 3. #GTCBerlin
Ralf Gulde retweeted
The real benchmark for physical AI is production.
300+ live systems across Europe and the US. One model learning from all of them.
Proud to be #5 in @Siftedeu’s AI 100. Thank you to everyone building this with us. 🦾
#SiftedAI100
Ralf Gulde retweeted
The physical AI startup founder pack
What am I missing here peeps?
🤖 Made with AI
With Cortex 1.6, robots learn from execution, not just outcomes.
Instability, retries, drifting forces.
The signal before failure.
At scale, that difference is decisive.
That’s the threshold we just crossed.
Most robot learning optimizes for success.
The real world breaks one step before.
With Cortex 1.5, we patch behavior where it fails.
Seconds of human correction.
Fleet-wide improvement.
Deployment isn’t the end.
It’s where learning starts.
Ralf Gulde retweeted
-the @SereactAI way-
The robotics world is slowly diverging from the previous consensus view that a research-only approach would lead to a ChatGPT-like breakthrough in the field.
Both recent pieces from @packyM and @oyhsu allude to a more step-by-step approach towards general robotic intelligence.
The reality is that real-world data is the only source where modalities, scale, and deployment distributions converge.
As such, exposing frontier robotics models directly to real production environments allows them to learn from live operations and align training and deployment distributions.
That's what then unlocks new use-cases (at the accuracy level customers demand), new embodiments, and the path to generalization.
Our portfolio company @SereactAI has been quietly building in this direction over the past couple of years, site by site deployment, scaling their real-world data loop while delivering outstanding customer value and unlocking new use cases and markets.
Not just demos. Real-world systems in production. All running on the same model, regardless of the use-case or robotic platform.
This is the way 🦾