← Sticker Wall AGENT · DETAIL

Recycleye: AI Vision + Robotic Arm Sorting for Waste, Acquired by CP Group after Reaching Approx. $7.1M ARR

Workflow: Waste on a conveyor belt continuously passes under cameras, where edge deep learning models identify the material, color

AGENT

Key Fields

FIELD STAMPS

IndustryEducation / Knowledge

RegionEurope（英国）

ScaleSME

ChannelOnline

🔧 Workflow

Waste on a conveyor belt continuously passes under cameras, where edge deep learning models identify the material, color, shape, and contamination level of each item within milliseconds (including detailed classifications such as food vs. non-food, aluminum cans vs. foil, black plastic vs. rubber). This is followed by automated sorting via a FANUC 6-axis robotic arm (up to 33,000 picks per 10-hour shift, 35 to 65 picks per minute) or QuantiSort high-speed air jetting. The input is raw waste flow, and the output is higher-purity sorted materials plus a real-time composition and purity dashboard. Plant quality inspectors act as human arbiters, conducting daily spot checks and adjusting strategies to form a continuously running closed-loop system.

🛠 Setup Requirements

Requires computer vision and deep learning R&D capabilities, along with industrial robot integration experience. The company partnered with FANUC to deploy 6-axis robotic arms, training recognition models based on its self-built WasteNet waste image database. Equipment can be retrofitted onto existing sorting lines and supports 24/7 operation. Founded in 2019 by Imperial College London graduates, the company went through years of R&D, pilot programs, and a $17M Series A led by DCVC. Following the 2026 acquisition, it has entered a commercial-scale replication phase. Too complex for an individual to build alone, it requires coordination between a skilled team and industrial partners.

🧰 Toolchain

🔧 WasteNet waste image database and deep learning models
🔧 FANUC 6-axis industrial robots (QualiBot)
🔧 QuantiSort 2.0 high-speed optical sorting and pneumatic ejection system
🔧 Real-time composition analysis data dashboard
🔧 Edge computing cameras and recognition units

💰 Revenue

ARR was approximately $7.1 million in 2023. The business model is hardware sales or leasing combined with annual subscriptions for AI software and analytics services (pricing undisclosed). Customer case studies show a payback period of about 1.7 years; material value doubles to triples after brick separation, and selling prices per ton rise significantly after fiber purity increases from 85% to 97%. Subscriptions and spare parts generate recurring revenue, which is accelerating post-acquisition through CP Group's channels.

💸 Cost

Costs are concentrated in FANUC robotic arms, optical sorting hardware, cameras, edge computing devices, model training, massive data annotation, as well as on-site installation, maintenance, and spare part services. Annual software subscriptions include continuous algorithm updates and analytics dashboards, creating a stable long-term operating expense. Exact figures are undisclosed, representing a hybrid cost structure of heavy assets plus software subscriptions.

⏱ Time Investment

Full-time team commitment covering algorithm R&D, delivery implementation, and customer success. Equipment operates automatically 24/7; a single robot achieves about 33,000 picks per 10-hour shift, replacing 1+ sorting workers per shift. Plant quality inspectors only need to conduct daily spot checks to maintain output purity, eliminating the need for continuous manual monitoring.

🚀 Getting Started

Individuals cannot directly replicate this industrial-grade system, but they can start by building a waste image classification demo using open-source computer vision frameworks. By cutting in through data annotation, quality inspection analysis, and small-scale pilot projects for recycling plants, one can accumulate industry knowledge and connections before transitioning to equipment integration, sorting solution consulting, or regional agency services. This industrialization path is more realistic than building robots from scratch.

🔑 Keys to Success

✅ Self-built proprietary WasteNet ultra-large-scale waste image database brings recognition accuracy close to the human eye, forming a hard-to-copy data barrier
✅ Software-hardware integration directly improves material purity and selling price, with quantifiable ROI and customer payback in about 1.7 years, simplifying purchasing decisions
✅ Deep strategic partnerships with industry players like FANUC and CP Group, leveraging established channels to rapidly penetrate European and American waste facilities
✅ Subscription-based SaaS plus analytics dashboards generate recurring revenue; more deployments lead to more data feedback and more accurate models, creating a powerful compounding effect

⚠️ 风险

⚠️ High industrial robot hardware investment and long sales cycles make cash flow dependent on financing; the eventual acquisition by CP Group also highlights the pressures of independent expansion
⚠️ Recognition accuracy is susceptible to changes in new packaging formats and waste composition, requiring continuous model updates to avoid compromising purity commitments
⚠️ Recycled material prices fluctuate with commodities, which may lengthen customer ROI calculation cycles; purchasing willingness can be suppressed when traditional optical sorting and labor costs are low

📌 Real Cases

📌 Ireland's largest recycling plant, Panda (processing 100,000 tons annually), deployed 4 robots on its aluminum line, replacing 1 worker per shift, achieving 98% uptime while improving purity and efficiency
📌 Partnered with MSS (a CP Group company) to complete over 23 Vivid AI optical sorting on-site installations across Europe and America
📌 Fiber sorting line purity increased from 85% to 97%, contamination rate dropped to 3%, line output increased by 10%, and recognition accuracy improved by 12%

Sources