Bring Down the Fences!
What we took away from TU Auto’s ADAS and Autonomous Vehicles event
Derrick Redding, COO
A couple of the major issues plaguing AV engineering teams right now is what to do about geographic and environmental fencing. Meaning, how can we increase the safety and reliability of AVs to let them loose on any road, in any condition, at any time?
Although many of the industry’s standing AV challenges are being addressed in numerous warm-climate pilot programs around the country, the environmental challenges outside of those pilots still remain. It’s time to bring down the geographic and environmental fences that are constraining those pilots. Here’s what we’ve learned.
At the recent TU Automotive ADAS and Autonomous Vehicles conference outside on Detroit, Carnegie Mellon University Associate Professor Philip Koopman described three categories of edge cases that AVs need to overcome to prove that AVs are as good as, if not better, than human drivers:
- Unusual Road Obstacles
- Strange Behaviors
- Extreme Weather
Current AV pilots test and simulate many conditions that include road obstacles and strange behaviors in geographically fenced areas where data can be gathered, analyzed, and used to help predict future obstacles and train vehicles to act appropriately. This is problematic as we scale AVs because geographically fencing vehicles means we’re merely gathering data and training vehicle behavior for one particular area over and over — like Groundhog’s Day: The AV Edition. We must certainly start R&D somewhere but our industry must also now look to identify how we can remove geographic fencing while ensuring safety. Additionally, a number of these AV fleet trials are avoiding other edge case categories altogether, by locating their trials in warm, sunny climates like Phoenix, Silicon Valley, and Miami. That is, the fleet operators are putting up environmental fences around the trials to reduce exposure to extreme weather. Furthermore, some fleets even program their vehicles to head straight to the garage when certain weather conditions like freezing rain hit.
It was interesting that, at the conference, I met folks from several companies that provide very accurate, precise weather information to AVs such as road surface temperature forecasts. I can see some real benefits to this type of information for AVs, however, this information is being used to avoid bad weather altogether and simply creating a more dynamic type of environmental fence. It may be expanding the boundaries of the fencing for fleets, but it doesn’t remove the constraints of this fence. For AVs to reach mass adoption, we must be able to increase vehicle uptime to close to 99.99%, which we can’t do when we avoid solving for an edge case completely.
One part of the environmental problem for AVs is that image recognition becomes more difficult in adverse weather. Another problem is how the environment can occlude the many and different types of perception surfaces on a vehicle such as cameras, LIDAR, and radar. These surface occlusions include snow, ice, bugs, salt spray, dirt, bird droppings, etc.
Long-term, the proven solution for clearing perception surfaces of occlusions is to provide heated washer fluid to these surfaces on an AV. Heated washer fluid is superior to cold fluid due to the lower viscosity of heated fluid, lower surface tension and higher solubility. This means higher spray pressures in cold weather, better lift-off of foreign materials, and better washing away. For many current AV fleets, the solution is to manually wipe the perception surfaces during downtime. This is not a scalable solution as AVs become the primary form of transportation.
At SEEVA, we’ve developed a camera and sensor cleaning system that can deliver heated washer fluid to any surface, in less than a second, to keep all these surfaces clear. This will enable AV fleets to eliminate environmental fencing constraints and increase vehicle uptime.
