(Again, reading up on ICRA)
- The basic goal is to find “half-steps” (I suppose half a gait, which they can do symmetrically for the other side?). At the start, end end of each the robot is statically stable and motionless
- Since planning in half-steps is half the size of the normal problem, they can do offline planning much more quickly than previously possible (1 hr)
- They use RRT* to get a feasible (no self-collision, no collision with obstacles) 2D sequence of half-steps (this planning takes about 14 seconds)
- Once RRT finds the raw sequence, it is smoothed into a fluid gait – they still need ot maintain feasibility, but this is pretty quick – about 12 secs.
- <Much of this is particulars to gaits so much is skipped>
- The half-step is broken into an upward and downward part