With collision avoidance, we can set a goal and some obstacles for each agent, so that the agent will "wander" towards the goal. In that sense, why do we still need to have pathfinding? Do you like to use which one in preference over the other or use a combination of both?
Thanks
Jack
Look at this picture (from wikipedia's A* article):
In this case, if your agent just "wander" from start to end, it won't find its path. But, as alvaro has said, if your world is free of big obstacle, then you can simply move towards it, dodging small obstacles. I would take a look at this then:
Create some obstacles and play around with the different searches http://qiao.github.io/PathFinding.js/visual/. Ask yourself how well you'd do if you could only know which direction the goal is from your current position and could only see a few feet in front of you. It should become clear why obstacle avoidance alone is not enough to get good pathing for certain obstacle arrangements.
