Good presentation at Boulder, if a bit dated (“10 cm of accuracy”), as six years is a while in high tech. It’s not (only) about not paying licensing fees to Google, although corporations (via their legal departments, initially, and by hardware, software, firmware engineers and cartographers after good guidance from them) have to pay attention to “how they use Google’s (map) data when making a map.” It’s actually a lot more complicated than that. As a professional cartographer for an AV company and specifically asking this client about how OSM might integrate into its maps OSM data (the answer from corporate was “as another tool, as a data source that can provide this company valuable mapping input”) with the company’s “cartography” (data, product, process…), it’s a LOT more complicated than that. It’s not (in my opinion), “a shame” that these are (necessarily) diverging. Conversations, dialog, potential integration strategies, questions about accuracy, update latency, safety…these continue.
Companies making AVs and especially making maps for AVs are teaching these robots a very specific set of aspects about “the road(s) around them.” They fit into a set of specifications that are amenable to being directly consumed by their vehicles — by design — as these data and processes “grew up together.” They are not the same (similar, but different) as what might be (or are) obtained by a crowdsourced public mapping project like OSM. I say that first-hand. And because I have signed NDAs, I shall not say more than that. These (mapping, cartographic) data diverge. For precise, technical reasons. Their creation is proprietary. They are simply not the same as OSM data and I think anybody can understand that, because they need to “feed” a particular set of inputs by an AI in an AV with realtime inputs and much sensorial and situational awareness (in a digital realm) with the AI itself being transported at speed. And most of the time, with human passengers, indeed as precious as it gets.
You might imagine how the concept of safety washes over huge amounts of, even the entirety of, these processes. Safety really must be a 100% constant concern of AVs.
Unless you are Google, “avoiding paying Google” (licensing fees) is really almost a small-to-vanishing concern regarding mapping / cartography for AVs. High-quality maps must be built with care, or they are not high-quality. “Using the map to cheat” (as Philipp said) is something both humans and AVs do, for similar reasons, in similar ways. Fixed objects CAN AND DO become different objects; think of a crosswalk being repainted.
I’m all for sharing (within the parameters of our ODbL), understanding that OSM might not “get back vision AI algorithms from AV companies,” and I’m OK with that. If AV companies are OK with sharing, I think there both is and can be symbiotic relationships in multiple directions. These seem to exist, even improve, especially as such dialog continues.