The future is a big place. There's no fundamental reason to expect this kind of work will not fall to automation, so it's more of a when question.
The trick is that the bulk of the assemblers will have to be replaced all at once to prevent those little accidents that damage robots.
For example, ship off-loading is nearly entirely automated in some sea ports with significant automation. As far as I know it wasn't done on a gradual basis, but by shutting down and building new. No long-shoremen would have been involved.
I expect if some robots were placed in a construction site alongside the workers they are partially replacing, that bad things would happen to the robots. Human beings are very competitive and have spent a long time learning about and practicing destroying their competition.
A balance to that is that robots have terrible spatial analysis skills, so if something about the environment changes, robots are currently crippled. In addition, robots have terrible weight to strength ratios, particularly if their power source is included. It's why exo-suits for humans have been promised since the 1950s and still haven't made it to useful application. This gives humans a huge flexibility edge that is hard to overcome.
It's the latter section that has kept robot groundskeepers from mowing baseball fields and then striping the baselines. People readily see the actions of the controls on the mowers and the stripers and, with some practice, just do it. Check the DARPA robot challenge to see where human-sized/capable robots are today.
The other characteristic that is presently insurmountable is that human brains have a prediction simulation section. I think it is what gives people their sense of self as they predict what will happen to themselves in the future. This is an area that AI hasn't even come close to yet and I think is between 50 and 500 years in the future. Even if a robot can analyze the space and obstructions around it, it still needs to rapidly make predictions about what will happen as it moves and as other things around it move. Think of how fast a bird flying through a forest has to account for every twig and branch as well as the movement of its own wings and the likely outcomes. Humans aren't quite as fast, but they consider the possibilities years to centuries into the future.
Robots still have trouble folding a towel because they can't predict the shape it will take when they grab a part of it and they can't predict which part they should grab next in order to get the shape they want.
I expect the most likely large industrial use will be ship yards where the various sections are welded together. Every part has a generally well predicted location so spatial requirements are minimized. The task of welding has already been clearly demonstrated and access to external power, already required by the welding equipment, would allow self-relocating robots to move to where the work is. Ship building has become seriously modular so the movement requirements are low. I guess it depends on whether the competition with humans is one they can survive.
So when? 50 to 100 years before all major outdoor construction will be by robots. A shorter prediction - little kids growing up now will have their grandkids marvel at it when they see a person push a lawn mower or a long-haul truck with a person steering it.
My favorite right now are these guys:
