The Hype of Geofencing for Autonomous Vehicles
While the strict definition of Geofencing means operating in a specific location the actual use is broader. In this space it means to cut down on use cases to incrementally develop and field this technology. That usually includes weather, complex road patterns, object types and movement and quantity of data that can be stored on board (mapping etc). As examples Waymo chooses a portion of Phoenix due its lack of precipitation in various forms, gridded and well-lit streets and lack of odd traffic patterns. Tesla chooses highway operations across the entire US and other places. The latter (and the lack of LiDAR) are why Tesla has more accidents and has needlessly killed the most people to date. (At least 6 of them in Tesla AP to date.)
While Geofencing can cut down on the work that needs to be done it is nowhere near the time and money saver people make it out to be. The problem of course being that the message these companies want to be sent is that the subset is significant. Thus, misleading people and providing false confidence. Geofencing is largely is hype for several reasons. All around Perception issues.
- The first involves accident scenarios. They still have to be learned for every road pattern in the area.
- The second involves detection of people especially around clothing pattern or finishes. Due diligence dictates you have to learn every object pattern that could be where you are. This means fabric patterns from around the world need to be learned for Chandler Arizona.
- The third involves fixed or environmental objects like buildings or trees. They can look different to the camera systems due to shadow differences. Shadows are often confused for solid objects, especially where there is no LiDAR. This means a common road pattern, like a 4 way intersection, may appear to be the same in many places but is not. Building and trees will be different enough to have to relearn a lot of the scenarios. (Heavy requirements on HD mapping by location support this as well.)
The solution here is to replace 99.9% of that public shadow/safety driving with aerospace/DoD/FAA level simulation and systems/safety engineering.
More can be found in my other articles
Autonomous Vehicles Need to Have Accidents to Develop this Technology
Proposal for Successfully Creating an Autonomous Ground or Air Vehicle
SAE Autonomous Vehicle Engineering Magazine — End Public Shadow/Safety Driving
