Revision history [back]

Hi moyashi,

if you set the dwa parameter to false, the local planner will use the Trajectory Rollout (TR) algorithm. The differences between the two are summarized here:

"DWA differs from Trajectory Rollout in how the robot's control space is sampled. Trajectory Rollout samples from the set of achievable velocities over the entire forward simulation period given the acceleration limits of the robot, while DWA samples from the set of achievable velocities for just one simulation step given the acceleration limits of the robot. This means that DWA is a more efficient algorithm because it samples a smaller space, but may be outperformed by Trajectory Rollout for robots with low acceleration limits because DWA does not forward simulate constant accelerations. In practice however, we find DWA and Trajectory Rollout to perform comparably in all our tests and recommend use of DWA for its efficiency gains."

I am using TR on the Pioneer robots in my lab, as I have found that it provides slightly better performance.

Some insightful comments on this subject, by the author of the base_local_planner package, can be found in these posts:

http://answers.ros.org/question/10035/questions-about-the-dynamic-window-approach-in-ros/

(http://answers.ros.org/question/10718/dwa_planner-vs-base_local_planner/)