Intro & scope
Use cases where AI perceives, decides and acts on physical machines: industrial arms, mobile robots, vehicles, drones, vessels, wearables and sensor nodes. Focus is EU deployment (Machinery Regulation, AI Act, GDPR, functional safety). Software-only analytics of camera data belongs in VIS; planning/scheduling solvers in OPT; predictive maintenance in PRED.
Scope boundaries between overlapping files: - ROBO-001 (fixed-cell picking) vs ROBO-014 (mobile unloading and mixed-SKU palletising): 014 covers trailer/container and pallet building, 001 bins, totes and conveyors. - ROBO-003 (programming interface) vs ROBO-013 (automatic path generation for welding/finishing): 013 is process-specific, 003 is generic. - ROBO-004 (VLA foundation models) vs ROBO-010 (simulation, training pipelines). - ROBO-016 is the safety layer for all machine use cases; it carries the high-risk AI Act reasoning (safety component of machinery). - ROBO-005/019/021/022 (road, rail, sea, mine autonomy) share a perception stack; 022 is geofenced private sites, 005 public roads. - ROBO-011 (TinyML sensors) vs ROBO-012 (smart home) vs ROBO-018 (exoskeletons): MCU-class vs hub-class vs wearable.
Table
| ID | Title | Pattern | Maturity | Path | HW tier | AI Act risk |
|---|---|---|---|---|---|---|
| ROBO-001 | Vision-guided bin picking & pick-and-place | ROBO.INDUSTRIAL | mainstream | A | HW1 | depends |
| ROBO-002 | Autonomous mobile robots (AMR) in intralogistics | ROBO.INDUSTRIAL | mainstream | A | HW1 | depends |
| ROBO-003 | Robot programming by demonstration & natural language | ROBO.INDUSTRIAL | early-adopter | A | HW2 | depends |
| ROBO-004 | Humanoid & general-purpose robots (vision-language-action models) | ROBO.FOUNDATION | emerging | A | HW3 | depends |
| ROBO-005 | Autonomous driving & advanced driver assistance | ROBO.MOBILE | early-adopter | A | HW1 | depends |
| ROBO-006 | Autonomous drone flight for inspection & delivery | ROBO.MOBILE | early-adopter | A | HW1 | depends |
| ROBO-007 | Agricultural robots (weeding, harvesting) | ROBO.MOBILE | early-adopter | A | HW1 | depends |
| ROBO-008 | Cleaning, service & hospitality robots | ROBO.MOBILE | mainstream | A | HW1 | depends |
| ROBO-009 | Construction robotics & automated site progress monitoring | ROBO.INDUSTRIAL | early-adopter | A | HW2 | depends |
| ROBO-010 | Simulation & digital twins for robot training (sim-to-real) | ROBO.FOUNDATION | early-adopter | B | HW3 | depends |
| ROBO-011 | Smart sensors & TinyML on microcontrollers | ROBO.EDGE | mainstream | B | HW1 | minimal |
| ROBO-012 | AI smart home automation | ROBO.EDGE | early-adopter | A | HW1 | minimal |
| ROBO-013 | Robotic welding & surface finishing with automatic path generation | ROBO.INDUSTRIAL | early-adopter | A | HW1 | minimal |
| ROBO-014 | Robotic truck & container unloading and mixed-case (de)palletising | ROBO.INDUSTRIAL | early-adopter | A | HW1 | depends |
| ROBO-015 | Autonomous ground inspection robots for industrial plants (legged & wheeled operator rounds) | ROBO.MOBILE | early-adopter | A | HW1 | depends |
| ROBO-016 | AI-based safety functions for machines & robots (person detection, speed & separation monitoring) | ROBO.INDUSTRIAL | early-adopter | A | HW1 | high |
| ROBO-017 | Robotic disassembly for remanufacturing & EV battery recycling | ROBO.INDUSTRIAL | emerging | A | HW1 | depends |
| ROBO-018 | AI-powered active exoskeletons for logistics, industry & care work | ROBO.EDGE | early-adopter | A | HW1 | depends |
| ROBO-019 | Automated train operation & AI rail obstacle detection | ROBO.MOBILE | early-adopter | A | HW1 | high |
| ROBO-020 | Last-mile sidewalk delivery robots | ROBO.MOBILE | early-adopter | A | HW1 | depends |
| ROBO-021 | Autonomous & assisted ship navigation (collision avoidance, remote-controlled vessels) | ROBO.MOBILE | early-adopter | A | HW1 | depends |
| ROBO-022 | Autonomous haulage & machine autonomy in mines, quarries and closed sites | ROBO.MOBILE | mainstream | A | HW1 | high |
Shared stack
Common layers: ROS 2 middleware; ONNX/TensorRT runtime; 3D perception (PointNet/SAM-class segmentation, open-vocabulary detectors); Isaac Sim/MuJoCo/Gazebo for simulation (ROBO-010); fleet manager and WMS/MES/CMMS integration via OPC UA/MQTT; certified safety PLC/safety I/O separate from AI (ROBO-016); OTA update and model registry with audit logging for AI Act traceability.
- HW0/MCU (ROBO-011, ROBO-018 firmware): Cortex-M/ESP32 class with TFLite Micro or Edge Impulse; gateway over MQTT/LoRaWAN; shared with PRED and ANOM sensor use cases.
- HW1 (edge, most ROBO cases: 001, 002, 006, 007, 008, 013-022): Jetson Orin/IPC with GPU (8-64 GB), ROS 2, TensorRT, safety laser scanners/LiDAR, local fleet server; one on-site GPU box can serve VIS-xxx camera analytics too.
- HW2 (ROBO-003, ROBO-009): single workstation/server GPU (24-48 GB, e.g. L4/RTX 6000) hosting a code LLM or VLM (Qwen/Llama-class, vLLM) for program generation and site-progress analysis; sim validation on same node.
- HW3 (ROBO-004, ROBO-010): multi-GPU node (4-8x H100/A100 class) or EU cloud GPU for VLA fine-tuning, RL and synthetic-data generation; robots run distilled policies on HW1 hardware.
- Beyond HW3: OEM-scale driving-model training (ROBO-005) needs large clusters; use EU-hosted cloud, not self-hosting.
Sharing rules: train once on HW3, deploy to HW1 fleets; reuse one perception/safety sensor set per site; keep personal data (cameras on public/ shared spaces, wearables) on-device with short retention for GDPR.