Open GiulioRomualdi opened 2 years ago
⚠️ we should test this PR on the robot to understand if the performances are improved
Based on my experience, the change in performance also on a custom example that uses those classes as the wbd does may be also a good indication in the increase in the performance.
I benchmarked the two implementations and I didn't notice any notable timing improvements
Let's see the bright side. If you have a simple executable it is much easier to profile the code with something like callgrind or flamegraphs
Removing myself from the reviewers, feel free to add me back if the PR is being worked on.
As a consequence of #141 I tried to improve the performances of the
ctrlLibRT
. Before explaining what I did I would recall that: ⚠️ we should test this PR on the robot to understand if the performances are improvedFirst of all, I tried to remove the copies from yarp to idyntree, thanks to this choice this code
becomes
so I removed two copies every time the joint state is filtered. A similar approach has been used for joint vel, joint acc, accelerometers and gyros, ft.
To achieve that I changed the interface of the
ctrLibRT
to considereigen::ref