Yes, I think this is another example that supports my thesis. If Waymo were driving "on rails" on the HD map, we would expect it to just stick dead center in the lane. The fact that it ping pongs in the lane is because the ML planner is given more freedom to maneuver. In this case, it looks like the ML planner wanted to pass the trailer. Presumably, the trailer was moving a bit slow and also the tree in the back was blocking the Waymo's sensors from seeing ahead. In any case, I think the ML planner wanted to get around the trailer. That ping pong behavior definitely looks to me like the planner was like "can I pass now? No. Can I pass now? No. Ok, can I pass now? No etc...". It saw that the right lane was a bike only lane so it was illegal to pass, but it kept looking for another opportunity to pass.
It was definitely awkward driving. Hopefully, it is straightforward to fix by just training the driving policy not to try to pass, especially if it knows it can't because the right lane is bike only. And yes, hopefully, we see Waymo retrain the ML planner to fix these issues. If I am right that Waymo is trying to be less dependent on HD maps and more reliant on the ML planner, then hopefully, once they fully "dial in" the planner, we should see a big leap in driving that is more generalized.