The interesting challenge AR has that VR does not is the backend infrastructure requirements. For most AR applications you are synthesizing content from relatio

The interesting challenge AR has that VR does not is the backend infrastructure requirements. For most AR applications you are synthesizing content from relationships in myriad external sensor systems continuously in real-time. VR requires a glorified game engine to a first approximation, a well-behaved Newtonian universe. There is a strong implication in AR applications that there exists a single logical model of physical reality behind the scenes, continuously updated from multimodal sensor and other data sources. Not only does this enable strong interoperability of individual user contexts in real-time, but giving everyone their own narrow model of physical reality will not scale in any case, the infrastructure needs to be shared. Any single user is seeing a contextually relevant and very tiny slice of that backend spatial data model. Companies cannot handle offline spatial processing of complex multimodal sensor data models at the required scales today, platforms don't exist with the necessary capabilities and architecture. AR is trying to do it on sensor feeds in real-time in a shared, event-driven environment at the same scales. The design and implementation of any such platform will be an extremely impressive engineering feat. Autonomous systems have a similar platform engineering problem. Think of it as machines requiring a real-time contextual view of their local physical reality synthesized from diverse sensor feeds, many of which are external. Very similar to the AR problem at a computer science level.