carlosjoserg
commented
9 years ago The low-level part is ready. You can use trajectory control in the joint position (with speed scaling) and stiffness (with reflex scaling), independently or simultaneously.
Looking forward to see some magic :tophat: :rabbit: :tophat: !
This is really hard in robotics. In most of the videos you see online, the pick-n-place tasks are not smooth, that is, dropping the object from the air and hoping that it arrives to the table in stand position, as in our recent video for the IROS submission. Visual-servoing can help in this situation, but also compliant helps t overcome uncertainties.
The KUKA LWR 4+ has the possibility to control stiffness, and the kuka-lwr package contemplates the capability to plan stiffness trajectories along with position trajectories.
What I envision is making the most of existing sampling-based planners with collision reasoning (not avoidance, as the classical approach). The reasoning translates in adjusting the combined trajectory (position and stiffness) according to the distance to obstacles and environment. Optimal stiffness for collision-free motions has been treated already in Centro Piaggio by the QBmove people. But, now it has to be combined with collision reasoning, in the sense that starting and ending motions might be in contact (typical pick-and-place operation). For instance be compliant for placing, open hand, and be stiff for retreat.
In the case of the Kuka, we can do it directly with independent controllers, since the stiffness is achieved via soft, but to illustrate the case, I resend this paper that @manuelbonilla sent me for mechanisms like the one in the QBs: http://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=4209770
/cc @enricocorvaglia