-
Notifications
You must be signed in to change notification settings - Fork 2
Development
At least one link is present in my URDF that wasn't in the original URDF. I added a link named 'base_footprint', rigidly attached to the 'base_link' frame, that represents the pose of the robot with the z component meaning the height of contact surface between the ground and the robot (as opposed to 'base_link' where z = when the robot is on flat terrain). This frame is relevant to the command messages going into the repositioning planner and grasping pipeline. Fields that mention base_link actually mean base_footprint, since it's effectively the root link of the robot.
The above configurations all provide a dummy transform between an external 'odom_combined' frame and 'base_footprint' to represent the odometry of the robot. It's a placeholder at this point (identity transform), but watch out for it breaking your TF tree.
For both base_footprint and base_link, the ROS convention is x = forward, y = left, z = up, even for navigation,...this apparently caused a LOT of problems last demo.
The reposition planner needs testing to see how well it works when MoveIt! is configured to plan in the robot's frame, rather than the world frame. There were some initial issues in the ik validity precomputation step. That should work now but hasn't been thoroughly tested. There is also an underlying issue with querying the arm planner/move_group, using the sbpl planing plugin, not correctly transforming the workspace bounding box into the diff-updated planning frame, that cripples computing base poses from non-preferred poses.
The collision model needs to be refined to account for the sensor suite in front of the robot and to have correct dimensions for the computer towers in the back of the robot. This needs to happen both in the URDF configuration and the bounding-spheres configuration for the sbpl collision checker for different planning groups.
The grasping strategy for the 3-finger Robotiq gripper needs a complete reevaluation. The current simple strategy is often too constrained and there are many cases where it can't come up with any reachable grasps wrt ik and self collisions.
Work began on a tool to analyze the workspace of the arm wrt ik and self collisions to function as a more accurate precomputation of valid base poses for grasping, but hasn't been completed.