Describe navigation/Roadmap/Brainstorm here.
Brainstorm
* Topological planning on global level * 3D lattice planning on room level * plugin vs ROS node model?
- swapping things out on the fly in the move_base node
- plugin model doesn't require a lot of change
* do we nedd to maintain both costmaps * how does the topological map get updated over time?
- this becomes a huge deal when we move away from global planning with something like navfn
* perhaps plan on room-level in the odometric frame * exploration? should we move into this * improving perception... integrating stereo, improving filtering * need some sort of interface for people to give goals if we move to the topological planner * some new trajetory follower to work with the dynamic planner * some generic architecture for a series of windows in which an arbitrary number of planners can operate
- slot different planners in
- goal pose comes in
- trajectory comes out
- plugin model for this?
- some way to decide which planners are getting cycles
- does each planner operate on its own thread?
- should we have a scheduler built in somehow
Short Term Goals (a few months)
- Working on lattice planning and a new trajectory rollout that will allow the robot to follow the paths from the planner closely.
- Improve navigation in narrow passages and doorways
- Improve navigation in cluttered environments
- Fix perception issues with 3D sensors
- Integrate topological map into the navigation stack
Medium Term Goals (some research)
- Some sort of planning for dynamic obstacles
- Need to be able to scrub the person out of the costmap
- Some sort of costmap that handles dynamic obstacles differently than static ones and tracks them
- A trajectory follower that takes dynamic obstacles into account
- Exploration and updating the map over long periods of time (SLAM online)
Long Term
- Extensibility and refactoring for arbitrary numbers and types of planners
