®
Since 2019, REBOTNIX has been developing Smart City applications that help cities become cleaner, safer and more efficient. The central challenge is the enormous amount of visual data. Depending on the use case, cities in North Rhine-Westphalia, Germany can generate up to 400 million images that need to be stored, understood, evaluated and translated into operational processes.
A traditional cloud architecture would be too expensive, too slow and difficult to scale. REBOTNIX therefore processes the data directly inside the vehicle.
Seven Models, One Vehicle
Up to seven highly optimized Vision Transformer models run in parallel and detect relevant conditions in public spaces. These models identify illegal waste dumping, damaged or obstructed traffic signs, construction site conditions, road damage and other safety-relevant events.
The key advantage is that the city no longer has to work with raw images, but with detected events that already include context, priority and operational relevance.
KINEVA Context Compression
For efficient data transmission, REBOTNIX uses KINEVA Context Compression. The technology reduces image data in a targeted way while preserving the relevant image areas needed for AI evaluation and documentation. Only what matters gets transmitted.
AGI Orchestrators as the Operational Layer
In the next step, the detected events are processed by AGI orchestrators. These orchestrators connect the results of the Vision Transformer models with context, priorities, responsibilities and operational workflows. A single detection is transformed into a controllable process.
Aligned with NVIDIA Jetson Thor
REBOTNIX's AGI architecture is aligned with NVIDIA Jetson Thor, a platform for physical AI, edge AI and industrial real-time applications. Jetson Thor is designed for high-performance AI processing at the edge: robotics, autonomous systems, sensor processing and real-time decision-making. NVIDIA outlines their vision for agentic AI in the physical world on their blog.
REBOTNIX uses NVIDIA NemoClaw as a software stack for autonomous agents and NVIDIA Nemotron as the AI model. Nemotron is fine-tuned by REBOTNIX at regular intervals on FP4 to continuously improve decision-making, prioritization and process orchestration.
Vision Transformers and AGI for the Self-Optimizing City with NVIDIA
The result is a self-optimizing platform for the city of tomorrow. Vision Transformers detect what is happening in public space. AGI orchestrators decide what should happen next. NVIDIA technology provides the compute power and software stack needed to bring this intelligence into vehicles and municipal operations at scale.
Benefits for Decision Makers
For city administrations, municipal operators and infrastructure leaders, the platform creates direct operational value.
Cities gain a technological foundation to move Smart City applications from pilot projects into daily operations.
