I found algorithms far superiour than me for 1-2d problems. (inverse pendulum problems, balancing a ball on a 2d plane and exactly controlling its position).
So problems probably are too many degrees of freedom, challenges like proper vision to plan ahead and steer, and in the shoving case also the security aspect (since there's a hard limit on how much force/acceleration a robot can use if it is handled by humans, and if you can shove it, it surely is).
In that sense, a skate-bot (without security limits) seems to be easier than a walking one with limits (which I assume you meant) to me since it takes down the degrees of freedom.
So problems probably are too many degrees of freedom, challenges like proper vision to plan ahead and steer, and in the shoving case also the security aspect (since there's a hard limit on how much force/acceleration a robot can use if it is handled by humans, and if you can shove it, it surely is).
In that sense, a skate-bot (without security limits) seems to be easier than a walking one with limits (which I assume you meant) to me since it takes down the degrees of freedom.