As the first wave of electric vehicles approaches end-of-life, the industry is racing to scale safe, economical ways to dismantle and recover high-value materials from lithium-ion batteries. Manual disassembly is slow, hazardous, and expensive; fully automated approaches have been more the subject of lab demos than industrial reality.
That’s why the EU-funded Recirculate project’s latest milestone matters: the team has completed the first phase of an AI- and robotics-based system that dismantles EV battery packs, moving methodically from pack→module→cell. The result tackles one of battery recycling’s biggest bottlenecks—throughput with safety—using computer vision, machine learning, and a purpose-built robotic cell.
What the System is and How It Works
Recirculate’s testbed is led by Centria University of Applied Sciences (Finland). A suite of purpose-designed end-effectors—including a vacuum gripper for lid lifting—plus a depth camera mounted on the tool, gives the robot the dexterity and 3D awareness needed to approach a high-voltage pack safely. The camera and models work in tandem: the AI identifies features; the depth sensor resolves the x-y-z coordinates with the precision required for fasteners and connectors.
A striking example is the lid itself: roughly 50 screws must be located and removed just to get inside many commercial packs. The team trained detection models to find every screw, extract its coordinates, and send them to the robot; once inside, additional models recognize connectors, wiring harnesses, and other components, and even infer wire orientation to choose safer, faster removal strategies. This pairing of perception and precise motion planning is what elevates the cell from a general-purpose manipulator to a domain-specific disassembly system.
Importantly, the system does more than pull screws: it also identifies battery packs even when QR codes or digital product passports aren’t present. In its current state, the model can recognize Ford and Tesla pack types with near-perfect accuracy, allowing the robot to automatically load the appropriate disassembly program—an essential step toward handling the dizzying variety of pack architectures in the field.
Why This Milestone Matters
The first phase was not a quick prototype. By the project’s account, it took about 18 months to develop the toolset and machine-learning models that enable robotized dismantling from pack to cell level.
That investment suggests the team is addressing real-world variance—different fastener patterns, occlusions, reflective surfaces, grime, and deformation—that routinely trip up lab-only solutions. As Centria’s lead notes, this is “one of the first working, real-world examples of battery disassembly using machine learning and robotics,” pushing the field beyond proofs-of-concept.
From a safety and economic perspective, the timing is important. A stream of reviews and technical studies over the past few years has argued that robotized EV battery disassembly is pivotal to scaling recycling with lower worker exposure to thermal, chemical, and electrical hazards while improving recovery yields. The newest systematic review of robotic disassembly consolidates that case, emphasizing the need for robust perception, adaptable manipulation, and process planning that can tolerate model drift and unknown pack variants—exactly the areas Recirculate is tackling.
Who’s Behind Recirculate
Recirculate is an EU-funded, three-year project, reported at €4.9 million in funding, and built around a multinational consortium that blends industry and research. Partners include Centria, Ford Otosan, DHL, CSEM (Switzerland), Eurecat (Spain), Probot, Minespider (blockchain/digital product passports), Libattion, and Eco Stor, among others. The mix reflects the project’s full-chain ambition: from robotic disassembly and pack identification through safe logistics and traceability for a circular battery economy.
What “phase one” achieved—and what comes next
Phase one delivers a working robotic cell plus the ML models to recognize and remove key elements—fasteners, lids, connectors, and wiring—so the robot can open a pack autonomously and begin structured teardown. The team also validated a battery-type identification model, demonstrating workflow automation from “what is this pack?” to “load the right sequence and tools.”
The next steps are about scale and generalization: expanding the training dataset to more pack families so the system can handle the diversity seen in real intake streams, and migrating from the lab cell to industrial environments with the appropriate safety, speed, and reliability certifications.
How the approach compares to the state of the art
If you map Recirculate’s design onto the field’s open problems, three choices stand out:
- Tool-centric perception: Mounting the depth camera on the end-effector reduces calibration drift between “what the AI sees” and “where the tool is,” a known failure point in early robotic disassembly. It also enables fine 3D localization of small features like screw heads—critical when tolerances are tight and surfaces are reflective.
- Part-specific ML models: Training separate detectors for screws, connectors, and cables aligns with the taxonomy of actions required for safe teardown (unscrew, unplug, unroute), mirroring best practices highlighted in recent academic reviews. This modularity can make it easier to update the system when new pack designs appear.
- Automatic pack identification: In real reverse-logistics pipelines, you frequently receive packs with incomplete labeling. A vision-based classifier that recognizes make/model can collapse minutes of human lookup into milliseconds and ensure the robot selects the correct sequence (e.g., hidden fasteners, glue lines, or “no-go” zones that vary by OEM). Recirculate’s early Ford/Tesla results indicate this is feasible at production-relevant accuracy.
The Bigger System Around the Robot
A single robotic cell won’t decarbonize recycling—but it’s a cornerstone for an end-to-end circular flow. That’s reflected in Recirculate’s broader scope: six pillars spanning state-of-health characterization, automated dismantling, repair/reuse, fast sorting, safe storage & transport, and blockchain-based battery passports & marketplace.
The presence of partners like DHL (for logistics) and Minespider (for digital product passports) hints at integration beyond the workcell—i.e., packs can be tracked, routed, and audited across the second-life/recycling landscape.
What Success Would Look Like
If Recirculate’s approach scales, recyclers and refurbishers could realize several concrete wins:
- Higher throughput and more consistent quality: robots don’t fatigue and can follow optimal sequences precisely. That improves yield and reduces damage to recoverable modules and cells. (A key goal repeatedly emphasized in the literature.)
- Lower safety risk: automating high-energy steps (opening lids, isolating busbars, unplugging HV connectors) pushes workers out of the line of fire and creates repeatable, auditable procedures.
- Better data for traceability: combining on-the-fly identification with a digital passport allows operators to record exactly what was processed, how, and with what outcomes—vital for compliance and secondary market confidence.
How This Fits into Europe’s Battery-Circularity Push
Europe has several research nodes advancing robotic disassembly—for example, teams at the University of Birmingham and collaborators in the Faraday Institution’s ReLiB program have demonstrated advanced robotic vision and 3D modeling pipelines for pack-to-cell disassembly. Recirculate’s progress resonates with this ecosystem, but shifts the emphasis to industrialization and integration within a multi-partner consortium.
Summary
Recirculate’s phase-one result isn’t just another lab robot unscrewing a lid—it’s a carefully engineered, perception-driven system built to cope with variety, ambiguity, and safety constraints in real EV batteries.
By pairing tool-mounted 3D vision, part-specific ML detectors, and automatic pack recognition, the project demonstrates a credible path to automated, pack→module→cell disassembly. If the team can extend model coverage across more OEMs and prove reliability on industrial lines, this work could become a reference architecture for European recyclers and second-life operators looking to scale safely and profitably.