Every minute, the world generates 2.4 million kilograms of municipal solid waste. Less than 20% gets properly recycled — not because the materials can't be recovered, but because human sorters on fast-moving conveyor belts miss up to 30% of recoverable material. A 12-person company from Barcelona's 22@ innovation district quietly fixed that. And they did it with a neural network that sees better than human hands can sort.
Sadako Technologies was founded in 2012 by engineers frustrated that recycling plants processing hundreds of thousands of tonnes per year were still relying on workers picking plastics off belts by hand. The margin for error was enormous — and costly. The technology to do better already existed. Nobody had applied it to waste. That gap became their business.
Their first robotic arm, Wall-B, was deployed at Ecoparc 4 in Barcelona — a municipal waste treatment facility managed by Ferrovial Servicios for the Barcelona Metropolitan Area. In its first operational phase, Wall-B recovered over 125 tonnes of PET containers per year: material that had been passing through existing machinery undetected, previously classified as reject. That single installation generated more than €50,000 in annual economic value from what was going to landfill. Ferrovial's results got the attention of Bulk Handling Systems (BHS), the world's leading manufacturer of material recovery infrastructure. BHS integrated Sadako's AI vision into their Max-AI platform — and a global rollout began.
What 80% of Recycling Plants Still Get Wrong — and What Sadako Fixed
Traditional waste treatment plants sort by colour and density using optical sensors. These systems work adequately in controlled environments. But municipal solid waste is chaotic: wet cardboard, tangled plastic film, broken glass, and contaminated bottles arrive together on the same belt at 3 metres per second. Conventional optical sorters struggle to distinguish a clear PET bottle from a translucent HDPE container at that speed and in those conditions. Human sorters compensate — but inconsistently, at high cost, and in physically demanding conditions.
Sadako's approach departed from the rule-based optical model entirely. They trained deep convolutional neural networks on waste-stream images as they actually appear in the plant: dirty, deformed, mixed, and partially obscured. Their models learn to identify PET, HDPE, aluminium, paper, and cardboard regardless of shape, deformation, or contamination — the way a trained expert would, but at machine speed. This distinction matters: most optical sorters are programmed with fixed rules. Sadako's systems keep learning from the live data stream, improving accuracy over time.
How the Technology Works in a Real Plant
Sadako operates two complementary AI systems. The first — Max-AI — combines physical robotic arms with deep-learning vision. Mounted above conveyor belts, Max-AI identifies and extracts individual recyclable objects at up to 1,500 picks per hour, classifying each by material type, colour, and recyclability in real time, then coordinating a pneumatic gripper to collect it from the belt without stopping the line.
The second system, RUBSEE, is a plant-wide AI monitoring layer with no moving parts. Up to 10 computer vision units are installed at different points along the conveyor network. Each captures a continuous image stream and identifies in real time which materials are flowing through each section — giving plant operators a live composition map of their entire waste stream. This replaces manual sampling, which previously required stopping the plant. RUBSEE generates automatic alerts when contamination spikes, when a material accumulates in the wrong stream, or when equipment performance drops.
The Numbers: What 36 Million PET Bottles and a 4.2-Month Payback Look Like
According to data published through the EU Horizon 2020 programme that co-funded RUBSEE's development, a standard installation with 10 monitoring units — costing €142,000 including hardware, installation, and first-year maintenance — enables a performance improvement of up to 20% per plant. For a mid-size municipal facility, that translates to 1,200 additional tonnes of material recovered per year, generating up to €421,200 in annual revenue from previously unrecovered recyclables. Investment payback period: 4.2 months.
At the global scale of their Max-AI deployments — facilities like GreenWaste Recovery in California and Repower South in South Carolina, both rated among the most highly automated Material Recovery Facilities in the world — the cumulative impact is measured in the tens of millions of units. One Max-AI installation near Barcelona has now surpassed 36 million PET bottles recovered since 2019. These are 36 million objects that would otherwise have been compressed into landfill, reprocessed instead into new packaging material.
The broader market validates the approach: the AI-powered waste management sector is growing at a CAGR of 14–19% and is projected to exceed $18.2 billion by 2033 (Market.us, 2024). Companies that have integrated robotic AI sorting report 40–70% increases in material recovery rates and 59–60% reductions in manual labour costs at sorting lines (IndexBox, 2026). The companies that moved first — like Sadako's clients — are now multiple recovery cycles ahead of competitors still relying on optical-only infrastructure.
Technology Stack
A full Sadako deployment runs on proprietary AI software on standard industrial hardware, with no specialised compute required beyond the vision units.
| System | Function in This Workflow | Access Model | Provider |
|---|---|---|---|
| Max-AI | Robotic sorting arms — 1,500 picks/hour, real-time deep learning vision | Commercial via BHS | BHS + Sadako Technologies |
| RUBSEE | 10-point plant monitoring — live waste composition map, automatic alerts | Commercial via Sadako | Sadako Technologies |
| Wall-B | AI robotic arm for mid-volume plants (100–500 t/year lines) | Commercial via Sadako | Sadako Technologies |
| Deep Learning Vision Module | CNNs trained on 1B+ waste images — identifies PET, HDPE, aluminium, paper | Proprietary (developed in Barcelona) | Sadako Technologies |
Who This Case Applies To
Sadako's model is directly relevant to any operator of mid-to-large scale material recovery facilities — particularly those still relying primarily on manual sorting or ageing optical equipment. The RUBSEE monitoring system is specifically designed for plants where a full robotic installation is not yet cost-justified: it layers AI-grade visibility over existing equipment without replacing it, with a 4.2-month payback that makes the ROI case straightforward for operations managers and procurement teams alike. Municipal consortia managing regional waste facilities, private waste operators seeking to improve material purity for secondary commodity markets, and industrial operators managing on-site waste streams are primary candidates. This is not a consumer or small-business solution — it requires access to an industrial waste processing facility with conveyor infrastructure.
Explore More AI Automation Case Studies at Sityos AI
Sadako Technologies is one of a growing number of companies turning AI into measurable operational ROI in industrial sectors. At Sityos AI, we publish weekly deep-dives into real automation case studies — with verified data, named companies, and results you can benchmark against your own operations.
Frequently Asked Questions
Do you need to replace existing sorting equipment to implement Sadako's AI?
No. RUBSEE is a non-invasive monitoring overlay — its vision units attach to existing conveyor structures without modifying mechanical infrastructure. Max-AI and Wall-B are more substantial installations but are typically deployed alongside existing optical sorters, not as replacements. Wall-B is specifically engineered for lines where conventional optical sorting is not cost-effective due to low throughput volumes.
Is Sadako's technology available outside Spain?
Yes. Sadako's AI vision powers Max-AI systems in facilities across the United States, Europe, Latin America, and Asia. Their partnership with Bulk Handling Systems (BHS) provides global commercial reach. The AI model development, training updates, and vision system improvements are all managed from their Barcelona office, which effectively serves as the AI backbone for all global deployments.
How long does a RUBSEE installation typically take, and what is the minimum plant size?
Sadako's EU project data indicates RUBSEE is optimised for facilities processing 100 tonnes or more per year, where manual sampling is the current alternative. A standard 10-unit installation can typically be completed within days without halting plant operations. For facilities at the lower end of that range, Sadako recommends a direct consultation to size the deployment correctly.
The waste sorting industry is one of the last heavy industrial processes still largely dependent on human hands. In 2026, AI-powered robots already achieve 99% material identification accuracy at 1,500 picks per hour — consistently outperforming human sorters on speed, precision, and uptime (IndexBox, 2026). Sadako Technologies built the vision that makes this possible. The gap between early adopters and laggards in waste treatment is now measured in tonnes recovered per year and millions of bottles diverted from landfill. The technology exists. The economics work. The only variable is when each operator decides to act.