Our goal is to shift recycling from hardware-led to software-led innovation
PICVISA is a Barcelona-based tech company specialising in AI, robotics and machine vision for automated material and textile sorting. Its advanced optical and AI-driven systems classify waste by composition, shape and colour, enabling efficient, high-quality recovery and supporting the circular economy.
In an interview with Fibre2Fashion, Sales Manager Silvia Gregorini explained how PICVISA’s multispectral, AI-driven ECOSORT platform outperforms manual and semi-automated textile sorting and highlighted that automation, data insights and flexible AI models will be essential to scaling global textile recycling as waste volumes surge.
What makes PICVISA’s automated textile sorting technology more efficient or reliable compared to existing manual or semi-automated methods?
Our advantage comes from combining hyperspectral NIR sensing, visible (VIS) imaging, and AI-driven computer vision in a single system. Manual sorting relies almost entirely on visual cues, which means operators cannot detect fibre composition accurately and performance naturally drops due to fatigue. With ECOSORT TEXTILE, every item is scanned simultaneously for its spectral signature which tells us its real composition and for its visual attributes such as colour, pattern, and shape. This removes the subjectivity and inconsistency of purely visual sorting.
Another key difference is flexibility. Our platform can be retrained or reconfigured very quickly: if a customer wants to group fibres differently, add a new pattern class, or separate garments by shape, the AI model adapts without needing to redesign the line. Manual sorting teams or semi-automated systems cannot match that speed of reconfiguration.
Finally, automation guarantees throughput and purity. Our ejectors operate in real time based on AI decisions, allowing high-speed, continuous production with repeatable purity levels.
In our case studies, we consistently see higher throughput and fewer mis-sorts compared to manual sorting, where fatigue and variability between operators can dramatically impact performance.
ECOSORT TEXTIL offers sorting by fibre, colour, and garment shape. Which of these is the hardest to automate, and why?
The hardest category to automate is garment shape. Fibre and colour are relatively well-defined: fibres can be objectively identified with NIR spectroscopy, and colours can be quantified through calibrated VIS imaging.
Shape, however, is much more complex because textiles are deformable objects. A T-shirt on the belt may appear folded, crumpled, inside-out, or overlapped with another garment. Unlike rigid products, fabrics change geometry constantly, which makes consistent shape recognition a real AI challenge.
We have developed deep-learning models trained on thousands of images of garments in different orientations and conditions precisely to address that. The progress is significant, but shape will always remain the most dynamic and unpredictable dimension of textile sorting.
What unique textile-specific data insights has PICVISA discovered that were previously invisible during manual sorting?
One of the biggest revelations is the true distribution of fibre compositions entering a plant. Manual sorting tends to overestimate ‘pure cotton’ and underestimate blends because humans cannot see fibre composition. NIR-based classification gives a far more accurate picture of what is in the waste stream.
We also quantify printed versus solid materials, which is extremely important because prints often cause errors in manual sorting. With AI, we can reliably separate solids, stripes, polka dots, checks, and other patterns; something operators simply cannot do consistently at scale.
Another insight comes from objective colour clustering. Instead of subjective categories like ‘blue/green,’ we produce precise shade groups and colour spectra. This data is extremely valuable for upcyclers and recyclers who need consistent inputs for dyeing, shredding or fibre-to-fibre processing.
How do the innovations presented at Ecomondo 2025 reflect PICVISA’s long-term roadmap for smart, AI-powered recycling?
At Ecomondo, we showcased how our technology is evolving beyond single-material applications. We demonstrated AI-driven innovations across textiles, plastics, and glass, highlighting our vision for modular, multispectral, multi-material sorting platforms.
Our roadmap is clear: move from dedicated, bespoke machines to flexible systems that evolve through software and data. ECOSORT embodies that deep learning models can be updated, new detection modules can be added, and sorting rules can be reconfigured quickly. The future of recycling will be software-led, and our work at Ecomondo illustrates that direction.
How does PICVISA balance processing speed with precision, particularly with challenging materials such as blended textiles, black plastics, or heavily contaminated waste?
We use a sensor-fusion strategy, combining fast VIS detection for colours and shapes with NIR/hyperspectral analysis for composition. This allows the system to make layered decisions: quick visual validation first, followed by precise spectral confirmation when needed.
Our AI classifiers also output a confidence score. Items with low confidence or ambiguous signatures can be diverted to a secondary analysis stream or manual check without slowing down the main line.
For particularly challenging materials like black plastics or heavily soiled textiles, we employ specialised preprocessing, tailored spectral models, and sometimes specific mechanical steps to reduce false negatives. This ensures we maintain both speed and purity.
How do you evaluate ROI for clients adopting AI-based sorting systems, and which KPIs best demonstrate long-term value creation?
We look at four main areas:
Throughput and labour optimisation: Higher tons per hour and fewer manual sorting hours provide immediate operational savings.
Purity and yield: The cleaner the product, the higher its value to mechanical or chemical recyclers direct revenue impact.
Reject management: Lower reject rates reduce rework and disposal costs.
Data-based value creation: Composition reporting, seasonality trends, and batch-level traceability unlock new commercial agreements, especially with fibre-to-fibre recyclers.
Our case studies show that combining labour savings with improved yields and higher sale prices for sorted fractions can significantly shorten payback periods.
Could PICVISA’s core technology expand into adjacent industries such as footwear, upholstery, automotive interiors, or medical fabrics?
Absolutely. Our system is inherently material agnostic: the multispectral hardware stays the same, and the AI models adapt to new categories. We already apply this approach across several material streams such as plastics, textiles, glass and extending into footwear or automotive textiles is a natural progression.
Each sector requires specialised classifiers. For example, coated materials, multi-layer composites, or flame-retardant fabrics but the ECOSORT architecture is designed precisely for this kind of adaptation.
What differences do you observe across global recycling infrastructures, and how does PICVISA adapt its technologies to local needs?
Global infrastructures vary widely. Europe is driven by strong regulation and mechanical as well as chemical recycling requirements. Other regions may prioritise throughput and low CAPEX due to lower labour costs or more heterogeneous waste streams.
PICVISA adapts through modular configuration like different sensor stacks and lighting setups, customisable belt speeds and ejector layouts and country-specific AI models trained on local waste streams.
We always begin with pilots and real plant data to ensure the system reflects the material reality of each market.
What role will cross-industry partnerships play in building a scalable textile recycling ecosystem?
They play a critical role. Textile circularity can only work if every part of the value chain moves in the same direction, and that means collaboration on several fronts.
First, the quality of the feedstock. Brands, collection systems, and reuse organisations need to supply cleaner, better-categorised materials. Automated systems like ours can deliver extremely high sorting precision, but only if the incoming textiles are traceable and reasonably prepared. Without partnership at the collection stage, it is impossible to unlock the full potential of advanced sorting.
Second, defining clear market standards. Mechanical and chemical recyclers need consistent input streams. Policymakers and brands need reliable recycled-content targets. Automation provides the data and repeatability to support both but only if everyone aligns on specifications. Partnerships help establish those shared expectations.
And third, building the right infrastructure. Scaling textile recycling requires real investment and smart plant design. Technology providers, waste managers, and industrial players must work together from the very beginning to de-risk projects, optimise layouts, and ensure that operations are both economically viable and future-proof.
At PICVISA, this collaborative approach is part of our DNA. We do not just install machines; we work with partners to understand their material flows, design solutions with them, test samples in our facilities, and refine the system together. The technology is the enabler, but the real progress comes from how well we integrate it across the industry.
How prepared is the global textile recycling industry to handle growing waste volumes, and how will automation become essential in scaling sorting and processing efficiency?
Today’s global infrastructure is far from ready. Europe alone produces millions of tonnes of textile waste each year, and recycling rates remain extremely low. Manual sorting cannot scale to the volumes or the purity levels that circular textile markets require.
Automation will be essential for three reasons:
Scalability: Automated lines can run continuously at industrial speeds while maintaining stable quality.
Purity requirements: Mechanical and chemical recycling processes demand consistent composition and colour purity that only sensor-based sorting can achieve.
Data and traceability: Automated systems generate composition reports, colour distributions, and batch histories that are critical for supply-chain transparency and future regulation.
In essence, automation transforms textile sorting from a labour-driven activity into an industrial, data-rich process capable of supporting true circularity.
