You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Right now, we work only with static data about the state of the world. We should be able to create a generic interface to handle detections to track them and have them included in the planning and control processes.
This could be done with explicit modeling (e.g. planner and controller step forward dt and we project the new time-variant pose of detections and uncertainty into C-space) or implicitly (e.g. we have some object we know is dynamic and we "smear" it in the costmap projecting its pose and uncertainty into the current costmap without being time-dependent)
e.g.
Dynamic avoidance by making the costmap virtually represent velocity vectors of obstacles for static planning
Dynamic avoidance by modeling dynamic obstacles and having a kinodynamic path planner (state lattice or RRT with t) search through costed space and vector space
The text was updated successfully, but these errors were encountered:
Is this still in development? If so, is the idea to first implement a dynamic obstacle detection (for example as a separate plugin) and then use in any local planner?
Integrations for dynamic obstacle data into DWB would be an appreciated contribution if you're interested. Basically it would involve creating a dynamic obstacle critic plugin to take in some set of dynamic tracks and using them and the change in time of the trajectory to score the paths.
Right now, we work only with static data about the state of the world. We should be able to create a generic interface to handle detections to track them and have them included in the planning and control processes.
This could be done with explicit modeling (e.g. planner and controller step forward dt and we project the new time-variant pose of detections and uncertainty into C-space) or implicitly (e.g. we have some object we know is dynamic and we "smear" it in the costmap projecting its pose and uncertainty into the current costmap without being time-dependent)
e.g.
t
) search through costed space and vector spaceThe text was updated successfully, but these errors were encountered: