Actually we allow people to hit walls, for the reasons you mention above, but we make sure that the robot slows down a little before hitting them. Basically we allow any movement but you can only go fast when you are in an open area and if the robot detects that you are going to barely clip an obstacle you barely miss it instead (thats the hope anyway)
Hey, I think my original post was totally misinterpreted. I was referring to the instance near the beginning where Arrington drove the robot forward, directly into Dan Casner, a few times. That mistake will almost certainly be repeated by everybody who uses QA, as they try to get a better view of the people they're talking to. So, I was commenting on the lack of a hard limit on how close one can get to objects directly in front of the robot. Will this be addressed in the final implementation?
I believe there might also be a ROS package that detects ankles (well, legs) in 2D scans, which could help.
-Benjie, anybots obstacle avoidance guy