Fundamental Autonomous Ground Vehicle technology has improved in recent years to the extent that they perform reasonably well on what would be considered good road conditions. But, considering driverless cars, they are having a difficult time navigating when conditions deteriorate. Technical improvements are needed to overcome certain deficiencies, such as when there's a blanket of snow or ice on the ground, or when dealing with vague human traffic direction, etc.
Even with fusion of any or all three of vision, lidar and radar, ground vehicles (when operating driverless) are finding it difficult to not only navigate, but to stay on the road. Hazardous road conditions such as driving snow and ice storms which affect both machine and human visibility. Railings along the road side help a great deal, but with rural conditions, where shoulders are soft to start with, or any sort of road markings/signage are few and far between, these are conditions that are proving very difficult. It seems that more elaborate decision implementations and applications of AI need to be done.
Some examples:
1) Another costly issue common with snow, is removing it. Snowy states spend quite a bit in snow removal through a winter season. Would the sensor suite or the applied AI be the weak link if a state wished to try using driverless trucks to do this?
2) Another scenario that happens often is dealing with road work, especially with humans directing traffic flow. For example, the other day while driving I came upon some road work where the traffic director was not watching me but made a slight hand gesture (a little wave about thigh high) instructing me to pass. I did drive through, but it made me think about how the AI unit in a ground vehicle would have recognized that as direction to continue and pass. I'm sure the vision system would've picked up the hand gesture, but right now I believe it might've had trouble understanding the intent.
3) Some other examples to consider when developing driverless technology for more than highway driving on nice days. It seems that trucks (pickups mainly) make up a large part of the US auto market, so when operating offroad, the conditions are less than ideal with narrow, unmarked and unpaved roads (class 6 types) typically being traveled. These roads may not be on the map, GPS can be difficult to pickup and the road can be in very poor condition. Avoiding animals in the road (in my state there are numerous moose collisions). Sometimes roads get redone and will no longer match what is on the map, this can prove difficult for the driverless car as well. At the moment, I'm not sure what the AI would need to consist of.
There are others I'm sure (probably a lot more). I believe some advances in applied AI, especially with decision making during the difficult scenarios mentioned and improved machine perception would help.
Your question is hard to find. Let me answer in the following way: Full autonomous driving is far away from serial production. So many issues are unsolved that it will take decades to have the full application range. Before a lot of steps are possible to introduce driver assistant systems to improve the safety of cars (where still the driver is in charge).
Automatic driving could be reached earlier with a teleoperated approach (still a driver beeing responsible but not sitting in the car). Have a look at my papers reagrding this topic.
- Badges
- Science topic
- Similar topics
- Engineering