# DEXOLYTA

**Adaptive AI for End-of-Life / Post-Crash EV Battery Disassembly**

The software brain on commodity robotic arms. Turning Europe's 2027–2031 battery recycling mandates from a compliance crisis into deterministic, scalable throughput.

---

## The Problem

Battery pack disassembly is the physical bottleneck of the EV circular economy.

- Published studies place it at **20–40% of total recycling cost**, executed manually under 800V shock, thermal runaway, and toxicity risk.
- Across hundreds of non-standardized pack variants entering the European waste stream, no manual workforce can scale to meet incoming volumes—and the technical workforce doesn't exist regardless of cost.

## Why Now — The Regulatory Forcing Function

The EU Battery Regulation (2023/1542) is binding law. Every European recycler has a hard wall:

| Deadline | Mandate |
|----------|---------|
| Feb 2027 | EV batteries must be removable by qualified professionals; digital battery passport mandatory |
| Dec 2030 | 70% lithium recycling efficiency required |
| End 2031 | 80% Li / 95% Co, Ni, Cu material recovery |

Volumes scale **6× over five years**: ~400k EoL packs (2025) → **>2.4M by 2030**. Compliance without automation is mathematically impossible.

## The Insight

Industrial robotic arms (UR, KUKA, ABB, Fanuc) are commodity hardware.

- The trapped value is the **perception + motion-planning brain** that handles structural variability safely.
- **We sell the brain, not the body.** Asset-light, software margins, hardware-agnostic.
- **Liability Isolation:** We deploy exclusively through certified System Integrator (SI) partners, isolating our operational liability to the software layer while partners handle physical cell integration and maintenance.

## How It Works

3D RGB-D + tactile perception → confidence-gated tool path planning → ROS 2 execution on any major arm → HITL teleoperation on edge cases → every teleop sequence feeds the training library.

The data moat compounds with every pack disassembled. We rely on four non-negotiable architecture choices:

1. **Synthetic Pre-Training:** Models are initialized on heavily augmented OEM CAD data—simulating real-world rust, warping, and thermal paste degradation—solving the cold start problem before day-one deployment.
2. **Confidence-Gated Execution:** Mechanical halt triggered automatically if confidence falls below 99.5%.
3. **Sovereign Self-Hosted MLOps:** OEM CAD data never leaves the customer's on-premise infrastructure.
4. **Productized HITL Teleoperation:** Engineered as a core feature, not a fallback, allowing immediate deployment while training the autonomous stack.

## Market & Unit Economics

- **2030 European TAM:** ~2.4M EoL packs/yr × €300–800 software value/pack = **€0.7–1.9B/yr**
- **Pricing:** €120–180k/yr per-cell SaaS license + per-pack throughput royalty.
- **Beachhead Sequence:** Switzerland (extreme labor cost + rapid pilot iteration via local innovation frameworks) → Germany (the core volume engine) → UK/France (regulatory pull).
- **Unicorn Path:** The perception stack extends horizontally to ICE vehicle teardown, white goods, and electronics dismantling—multiplying TAM 3–5× with the exact same data infrastructure.

## Competition — Honest Delta

| Player | What they do | Delta |
|--------|--------------|-------|
| Li-Cycle, Redwood Materials | Vertically integrated recyclers | We are horizontal software; we sell to them. |
| Recirculate (EU-funded R&D) | AI/robotics disassembly project | Pre-commercial; we ship product, not papers. |
| READY Robotics | Brand-agnostic industrial robot OS | General-purpose; we are battery-domain specialized. |
| Academic cells | Research prototypes | Not productized; we own the deployment layer. |

## Defensibility Against Pack Standardization

Future EU regulation pushes design-for-disassembly, threatening the variability moat.

**Counter:** The 2015–2030 European vehicle fleet is already locked. Legacy pack variability persists in the waste stream through **2040+**. We capture the long, profitable tail—and our data flywheel makes us the default brain when standardized packs do arrive.

---

## Papers

- https://arxiv.org/pdf/2606.06493v1
- https://arxiv.org/pdf/2606.07512v1

## Pilots / LOI / MVP

- https://librec.ch/en/
- https://www.srf.ch/play/tv/10-vor-10/video/die-idee-e-batterien-recycling?urn=urn:srf:video:6c822ba7-6cef-4d22-90be-a52a8956a3fb

## Competitors

| Company | Tier | Primary Use-Case | HQ / Location | Why Relevant |
|---------|------|------------------|---------------|--------------|
| WasteFlow SA | Startup / Spin-off | AI-driven sorting & recycling (vision copilot) | Ecublens (VD), EPFL CVLab | Real-time vision detects/classifies waste type, mass & hazard inside existing recycling plants; +50% efficiency claim; live pilots (Retripa) |
| UpCircle Technology | Startup / Spin-off | AI recycling analytics / material intelligence | Zurich, ETH RSL spin-off (2022) | Computer vision quantifies infeed material quality of waste bales for recyclers/paper mills; replaces subjective human grading |
| Almer Technologies (acquired) | Startup | AR remote maintenance & guided repair | Bern | Lightweight AR headset + remote support for industrial maintenance/repair; clients incl. Coca-Cola, Ford |
| Augment IT (Netcetera group) | SME / Integrator | AR visual guidance for maintenance & inspection | Zurich | Inspect AR + remote support overlay step-by-step 3D instructions for technicians on machines/plants |
| afca. ag | SME / Integrator | AR maintenance & inspection instructions | Switzerland | AR work instructions on HoloLens/smartphone for maintenance, inspection, training |
| Akatech SA | SME / Integrator | Industrial machine vision & inspection | Ticino (since 1995) | Custom CV/quality-control & anomaly detection systems; could extend to disassembly/sorting vision |
| ABB (Robotics) | Established / Enterprise | Robotic disassembly & sorting automation | Zurich HQ | Robots + machine vision power e-waste disassembly (Hiro Robotics partnership): unscrewing, sorting, hazardous handling |
| Bühler Group (SORTEX) | Established / Enterprise | AI optical sorting (food/grain today) | Uzwil (SG) | Deep-learning full-color/InGaAs optical sorters; sorting-tech leader that could extend into recyclate streams |
| Immark AG / Thommen Group | Established / Operator | E-waste & battery sorting operator (AI user) | Regensdorf (ZH) | Largest CH e-waste + new automated battery-sorting line; pilots SENS AI on conveyor — an internal alternative & key customer |
| SENS eRecycling / SWICO | Established / Industry body | E-waste data & take-back; AI detection project | Switzerland | 'SENS AI' increases transparency of small-appliance flows; gatekeeper of recycler relationships & data |
| TOMRA Recycling | Potential entrant (intl.) | Sensor-based / NIR + AI sorting | Norway (active in CH) | Dominant sensor-based sorting hardware in CH plants; AI-enhanced sorting; could push software layer |
| ZenRobotics / Terex | Potential entrant (intl.) | AI robotic waste sorting | Finland/US (deployed in CH) | AI waste-sorting robots deployed by Eberhard Group in Switzerland — direct overlap with disassembly/sorting |
| RealWear / TeamViewer | Potential entrant (intl.) | AR wearables + remote assistance | US/Germany (owns Almer) | Now owns Swiss Almer; TeamViewer Frontline is a scaled AR remote-assist platform that can dominate CH repair-guidance |
| Empa + KYBURZ | Research / Pipeline | Robot-assisted battery disassembly/recycling | Dübendorf / Freienstein | Robotic Li-ion battery disassembly plant & 'Battery Expert System'; talent & IP source, future spin-off potential |
| ETH RSL (MARC / ARC) | Research / Pipeline | Autonomous robotic sorting research | Zurich | Robotic sorting container + autonomous detection research; feeds startup pipeline (UpCircle origin) |