75,024 Seats.
4 Hours.
Zero Compromise.
Robotic precision cleaning for the Allianz Arena — powered by NVIDIA Physical AI full-stack platform and dual-arm robotic technology with fixed multi-camera vision.
After the Game, the Real Work Begins
After every match at the Allianz Arena, 75,024 seats must be cleaned within a tight 4-6 hour window. The three-dimensionally curved HDPE seats with seat surface, backrest, and armrests present an enormous logistical challenge.
Labor Shortage
Difficulty recruiting 150-200 cleaning workers for night shifts after every match
Inconsistent Quality
Manual cleaning varies greatly — no standardized quality control possible
High Operating Costs
EUR 30,000-50,000 personnel costs per match day, over EUR 1.5M annually
Complex 3D Geometry
Three-dimensionally curved seat shells require adaptive cleaning movements
Dual-Arm + Fixed Vision Architecture
Our autonomous cleaning robot features a tracked platform with two specialized 6-axis robotic arms on the seat-facing side (steam cleaning + brush-suction) and a fixed multi-camera vision module on the aisle side for real-time AI inspection and guidance.
Arm-S: Steam Cleaning
High-temperature steam jet at 140-160°C for deep cleaning and disinfection
Arm-B: Brush & Suction
Rotating brush with integrated vacuum suction for instant drying
Fixed Vision: AI Inspection
Fixed multi-camera module with stereo depth sensing for stain detection and arm guidance
BUNKER Pro 2.0 Tracked Platform
The tracked platform from AgileX Robotics provides the perfect combination of load capacity, stair-climbing ability, and precision navigation for stadium terrace operations.
Robot positions on aisle steps, arms reach into seat rows. No entry into narrow 350-450mm row gaps.
Embodied AI Perception System
NVIDIA Jetson Orin Nano Super as the central computing unit, combined with a fixed multi-camera vision module (2x Intel RealSense D435i) and LiDAR for real-time 3D environment perception, SLAM navigation, and AI-based stain detection.
NVIDIA Jetson Orin Nano Super
Intel RealSense D435i x2
Livox Mid-360
ROS2 Navigation Stack
Real-Time 3D Mapping & Path Planning
ROS2 Nav2 stack with SLAM Toolbox creates real-time 3D maps of the stadium terrace, enabling autonomous navigation between seat rows with centimeter-level precision.
Dual PiPER 6-Axis Robotic Arms
Two AgileX PiPER lightweight 6-axis robotic arms mounted on the seat-facing side, each equipped with specialized end effectors: Arm-S for steam cleaning and Arm-B for brush cleaning with vacuum suction. A fixed multi-camera vision module on the opposite side provides real-time guidance.
Full-Stack AI Development Pipeline
Leveraging the complete NVIDIA Physical AI ecosystem — from DGX Spark training clusters to Omniverse digital twins, Newton physics simulation, Cosmos synthetic data, Isaac Lab training, and GR00T foundation models — for end-to-end robotic intelligence.
Compute Infrastructure
128GB unified memory, 1 PFLOP FP4 — local training for confidential BIM data
67 TOPS edge deployment, upgrade path to Jetson Thor (2070 TFLOPS)
Virtual Training Environment
Digital twin of Allianz Arena from BIM/CAD + LiDAR point cloud
Differentiable physics: steam spray, brush contact, vacuum dynamics
5M synthetic stain images, 100K virtual stadium configurations
Training & Foundation Models
512 parallel environments, 5M training rounds, ~72h training cycle
Vision-Language-Action model: see stain → understand task → execute trajectory
>95% accuracy across beer, ketchup, gum, mud, cola stains
Validation & Deployment
Simulate 40 matches (34 Bundesliga + 6 Champions League) in digital twin
Jetson Orin Nano → Jetson Thor upgrade path for fleet scaling
Superior Efficiency Through Pipeline Cleaning
| Metric | Manual | RockSmart | Unit |
|---|---|---|---|
| Staff per Match | 150-200 | 25 + 3 | Workers / Robots + Operators |
| Cost per Match | 30,000-50,000 | 2,500 | EUR |
| Annual Cost (50 Matches) | 1,500,000-2,500,000 | 640,000 | EUR |
| Cleaning Consistency | Variable | Standardized | |
| Data Collection | None | Complete |
Industry-Proven Components
Every component comes from leading manufacturers with proven reliability. Prototype cost: CNY 160,518 — Mass production (100 units): CNY 95,000 per unit.
BUNKER Pro 2.0
PiPER Arm x2
Jetson Orin Nano Super
RealSense D435i x2
Steam Nozzle
Roller Brush
Vacuum Motor 500W
Kamoer KPA200
20L Pressure Tank
48V LiFePO4
Livox Mid-360
Karcher SG 4/4
Role Positioning
System design, HW/SW integration, AI training, deployment & operations
Physical AI full stack: Jetson + Omniverse + Isaac Lab + Cosmos + Newton
BUNKER Pro 2.0 tracked platform + PiPER 6-axis arms
RealSense D435i depth cameras + Mid-360 LiDAR
Requirements, venue data, deployment acceptance
12 Months to Deployment Readiness
Single-arm cleaning validation on test seats, DGX Spark cluster setup
Dual-arm + fixed vision integration, Omniverse digital twin, Isaac Lab training
200-seat simulated terrace, 48h endurance, GR00T N1.7 deployment
5 robots at Allianz Arena, one section ~5,000 seats
CE certification, 25 robots, 75,024 seats full coverage
Ready for a Demonstration?
We invite you to experience our prototype in action. Contact us for a personal presentation or a visit to our R&D lab in Shenzhen.