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
Our current object placing strategy, which is implemented in the place action, is "blind" in the sense that the robot just moves to the desired placing pose and releases the object there. This is however suboptimal because it can lead to unexpected failures (in particular, light objects may fall down if released from too high above, while fragile objects might even break).
Suggested solution
To improve this, we should change the placing workflow so that it is performed in a closed loop. I particularly suggest modifying the placing strategy so that the robot first goes to a desired placing pose (as done in the current implementation) and then moves down until the object touches the surface.
The primary challenge here would be finding a way to reliably measure the impact the object makes with the surface, while also avoiding false positives from potential impacts the arm may make with the surface. As a start, we could see if the HSR's force sensor is already good enough for this purpose.
The text was updated successfully, but these errors were encountered:
Problem description
Related to #219
Our current object placing strategy, which is implemented in the
place
action, is "blind" in the sense that the robot just moves to the desired placing pose and releases the object there. This is however suboptimal because it can lead to unexpected failures (in particular, light objects may fall down if released from too high above, while fragile objects might even break).Suggested solution
To improve this, we should change the placing workflow so that it is performed in a closed loop. I particularly suggest modifying the placing strategy so that the robot first goes to a desired placing pose (as done in the current implementation) and then moves down until the object touches the surface.
The primary challenge here would be finding a way to reliably measure the impact the object makes with the surface, while also avoiding false positives from potential impacts the arm may make with the surface. As a start, we could see if the HSR's force sensor is already good enough for this purpose.
The text was updated successfully, but these errors were encountered: