Zalando Invests $6M in Sereact’s Physical AI: Inside the $116M Series B and the Robot That Handles Your Returns

Why Zalando Invested: Returns Are a $6M Problem Worth Solving

Zalando contributed roughly $6 million as a strategic investor, lifting the round from the $110 million led by Headline in April 2026 . The significance is twofold. First, the investment solves Zalando's own acute pain point: every returned item must be opened, inspected, folded, and restocked—a process that Sereact's dual-arm station is built to handle end-to-end . Second, Zalando's participation serves as a powerful validation signal. Zalando is not a passive financial backer; it is a strategic customer-investor, putting its own operational weight behind the physical AI thesis .

The Cortex Brain: One AI Core for Any Robot

At the heart of Sereact's platform is Cortex, a vision-language-action (VLA) model that functions as a universal AI operating system for robotics . Cortex runs across multiple hardware embodiments—single-arm picking cells, dual-arm returns stations, humanoid robots, and fixed cells—without needing to be retrained for each form factor . Because it plans in visual latent space rather than robot-specific joint commands, a skill learned on one robot type transfers directly to another .

Dual-Arm Return Automation: The Solution to Fashion's Biggest Headache

Sereact's dual-arm station is purpose-built for the returns workflow . It autonomously opens packages and shoeboxes, extracts items, inspects for damage or anomalies, folds clothing (including jeans), scans barcodes and labels, and routes goods back into inventory, to refurbishment, or to disposal . The system mimics the judgment of a skilled warehouse associate but operates consistently across every shift and every brand . This directly targets what has been the fashion industry's most stubborn logistical bottleneck: the unpredictability and labor intensity of processing returns .

The Data Flywheel: Over 1 Billion Real Picks

Sereact's key structural advantage is its closed-loop data flywheel . More than 200 Sereact systems are live across Europe with customers including Daimler Truck, Mercedes-Benz, BMW, PepsiCo, Austrian Post, and Rohlik Group . Those systems have completed over one billion real production picks—not simulated or lab data, but actual picks from live operations . Every pick, whether a success, a failure, or a recovery, is captured with synchronized observations, robot state, and force feedback, then fed back into a centralized model that is continuously retrained and redeployed across the fleet . The fleet already achieves a reliability ratio of roughly one intervention per 53,000 picks requiring remote human help .

Cortex 2.0: From 'Try and See' to 'Plan and Try'

Cortex 2.0, the next-generation model that the Series B funds are specifically allocated to scale, augments the current VLA model with a learned world model that simulates physics and object behavior . Instead of acting and correcting mistakes reactively, the robot now generates multiple candidate future trajectories from its current state, scores each against the world model for stability, risk, and efficiency, and commits only to the best-scored branch—updating in real time as the scene changes . This paradigm shift from "try-and-see" to "plan-and-try" is critical for high-stakes tasks like assembling components under tension, fragile-item placement, and kitting, not just picking . The planning compute is tunable per task, spending more foresight where failure is expensive and less where recovery is cheap . The Series B capital is also funding Sereact's international expansion, including a new office in Boston .