Kylemonger makes a good point about the construction industry having to tighten up how they demarcate construction sites. In fact; I said as much myself. Thing is; there already is standard signage
along with real barriers, standards for flagmen etc etc. And; as I pointed out in my original post, even human drivers screw up regularly in those situations. Replacing pylons and barrels with say Jersey barriers would drastically increase the cost of road repair, adding utility connections to buildings and so on. The reason we use the barrels and pylons is because it is fast, easy and flexible. When doing certain types of road repair, it is common to have "rolling sites", where the workers progress down the road at a slow speed (crack sealing is a common example) while the traffic control guys grab pylons or barrels from the back of the site and shuttle them up the front, arranging them to extend the leading edge of the site. This is usually done by having the foreman driving a pickup full of collected pylons up to the front of the site and dropping them off for the forward flagman to arrange as he or she goes.
Doing that with Jersey barriers or crossbucks would be a lot slower and more expensive. Moving Jersey barriers requires heavy equipment, can only move a few at a time and forces traffic behind it to move even slower.
which brings me to Coren22's post: It is a pretty strong rule that construction sites and accident scenes must disrupt traffic as little as possible. And sometimes it just isn't feasible to close an entire road. I remember one job site where it was a two lane city street and the crews needed to dig a large trench across both lanes and have that trench open for several days to allow for new water mains, gas lines and so on. But the city refused to give us permission to totally close the road because it was a preferred route for tour buses to get to the bottom of Clifton Hill. In addition, being a major tourist area (Niagara Falls Canada) we were not allowed to leave excavations open after we shut down at the end of the day either. So; what we ended up doing was closing one lane, flagging buses in the usual alternating style on the remaining lane while work was done in the closed lane. Then the first lane would get filled in, and everybody would swap sides so work could be done on the other side. This slowed everything down tremendously. What should have been three or four days of open hole turned into ten days. (The craziest part? you aren't supposed to put removed material back into the hole because it doesn't pack or settle predictably, so every time we emptied a hole, the burden was taken off for fill and fresh gravel was dumped in the hole. )
In the case of accident scenes, there would already be traffic on that road that couldn't be rerouted. Sure, you can close a highway at the first off ramp behind the scene, but there is almost certainly going to be a certain amount of traffic already past that point. What's an autonomous vehicle supposed to do then? Without a driver on board, it can't proceed and it certainly can't be allowed to just park and wait either.
My own idea is to set a standard for "follow me" vehicles, like they sometimes use at airports. Any job site or accident scene gets two or more "Follow Me" vehicles assigned to it, with a human driver in it. All vehicles, autonomous or piloted, get required to follow it until it sends a certain signal (coded IR light perhaps?) and pulls over out of the way. At which time the traffic can proceed normally. Autonomous vehicles are already good at playing "follow the leader", so this would be a pretty easy system to implement.