While it is, of course, possible to implement animations within an AI architecture, it would be a poor methodology for a number of reasons.
One such reason would be an agent that chooses to walk but cannot for some reason. Another reason would be needless complexity (many decisions would have to implement an exact-copy of an animation node because they all result in the same sort of behavior). Most importantly, however, is that your pathfinding will not be handled in the behavior tree, and the pathfinding algorithm is what will actually implement agent movement. Remember that the AI is only meant as a means of decision making, not implementing those decisions in the world.
A much more appropriate solution would be to tie the walking animation to the method that adds velocity to the character, which would be the pathfinding algorithm. This would implement the walking animation, or could implement a different animation for a different form of movement (running, crawling, teleporting, flying). It would be able to handle obstruction, difficulty in determining a path, and more all of which should interrupt any walking animation occurring. And it would simplify your behavior tree significantly, which is incredibly important for readability and debugging.