In generating this data, I chose all the open-source frameworks, packages and libraries I know of. I left out proprietary software, like MATLAB (which gained 173 mentions). The names I searched for include those of component libraries, like FCL, simulation libraries (like Gazebo), several planning libraries (OMPL, SBPL, CHOMP, etc.) and complete open-source frameworks, like MoveIt!, OpenRAVE and ROS. The ROS-Industrial consortium is also included in the list. The list is, admittedly, biased towards the ROS ecosystem. The numbers were generated by searching for the number of papers where the names appear in the proceedings of the conference. I counted each paper where a name appears as a single mention even if the name was mentioned multiple times in the same paper. The search was performed using the search functionality in the proceedings so it is likely that some mentions may not be accurate.
The first graph shows the number of mentions for each software package. It is clear that ROS has had a huge impact, gaining mentions in 100 papers (of a total of about 930 accepted papers). Vision frameworks like OpenCV and PCL also garner a large number of mentions. Gazebo seems to be the most popular open-source simulation framework while the ROS-based planning frameworks and packages, like MoveIt!, OMPL, etc. also gain more than 10 mentions. Users seem to be about equally split between MoveIt! and OpenRAVE. Overall, it is clear that open-source software is having a big impact on robotics research in manipulation.
I also looked at which robots garnered the most mentions in manipulation research. In most cases, I looked for the names of the robots themselves, e.g. the PR2. In other cases, I chose to look for the names of vendors, e.g. KUKA. The list includes robot arms, hands, humanoids and mobile manipulation systems. Note that I am only counting the mention of a particular robot - this does not necessarily mean that the robot was used in the research.
Other industrial robot platforms, in general, do not rank high on this list. It is interesting to ask why. Is it safety and setup issues? The lack of a common, open API (an issue only now being addressed through the efforts of the open-source community)? The difficulty in having to integrate multiple components (grippers, sensors and more) vs. fully integrated systems like the PR2? Or is there a need for better collaboration between academia and industrial robotics? It is clear that we have reached a turning point in robotic manipulation, with the promise of new features, new capabilities and new robots. What seems to be missing is an overall software framework that brings all this together the right way - all the pieces of the puzzle are there, somebody just needs to put them together the right way.