Traffic lights are (in my opinion) a 'solved problem' - they are ubiquitous.
A humanoid robot would be many orders of magnitude harder. You have the mechanics of the body rotation, arm movements etc, then the sensors to detect current traffic patterns, the algorithm development to make it all work and the power supply issue to power it for however long it takes.
This is probably why you have traffic lights on many intersections but I've not seen anything like what you suggest (i.e. ASIMO and the like) as the cost/benefit (over hiring a person) is just not worth it (at least at the moment).
Susan