Modern frontend applications are increasingly connected to the physical world - but most frontend architectures are still designed as if data were clean, ordered, and reliable. In autonomous transportation, that assumption breaks immediately.
This talk is a deep dive into building a production-grade real-time control room for autonomous vehicles, where a React-based web UI is responsible for monitoring, visualizing, and supporting human intervention in live vehicle operations. The system integrates real-time telemetry, MQTT/Kafka event streams, 3D map and sensor visualization (Three.js), and video feeds - all under strict latency and correctness constraints.
I will share concrete architectural decisions, failure modes we encountered in production, and the patterns that allowed us to keep the UI responsive, truthful, and safe under unreliable networks and out-of-order data. We will look at why common frontend state patterns fail in real-time systems, how rendering performance directly impacts operational safety, and how to design UI feedback when the underlying system is only eventually consistent.