The PRM was very effective in solving the pac-man problem. As you can see, it came up with a very direct graph. The PRM took 19.65 sec to construct the PRM, and only .06 seconds to connect the initial to the goal!
This page will show some of the different available motion planning algorithms, and compare and contrast them to ours.
The PRM was very effective in solving the pac-man
problem. As you can see, it came up with a very direct graph. The PRM took
19.65 sec to construct the PRM, and only .06 seconds to connect the initial to
the goal!
The RRTConCon was effective in solving the
pac-man problem. You can see that a reasonable solution was generated. It took
the RRTConCon about 3.8 sec to come up with a solution to the problem.
The RRT was unable to solve the pac-man problem. It
took over 220 sec before giving up.
The RRTGoalBias method was not able to solve
the pac-man problem. I let the algorithm run for over 180 sec before giving
up.
The RRTStar algorithm didn't even come close to
solving the pac-man problem. As you can see it didn't make it around any of the
maze corners. I stopped trying after waiting about 210 sec.
Our Randomized Potential Field algorithm was able to
successfully solve the pac-man problem. You can see that we did have a little
trouble with local minima, but the bounce walk function was able to eventually
make it through.
It took the RPF a total of 6.92 sec to solve this problem, and it got
stuck 20 times. There were a total of 559 nodes in the path from init to
goal.
This is similar to our last attempt, except
we turned the path smoothing on. It again took about 7 seconds to solve the
problem, but only got stuck 6 times, which is a random event, not a function of
the path smoother.
The path smoother took 2.3 seconds, and was run with smoothingIterations=3.
When you play the animation of the robot moving through the world, the path is noticeably
smoother, and less jerky.