VisGraB: A benchmark for vision-based grasping of unknown objectsVisGraB is a benchmark for vision-based grasping of unknown objects. The benchmark includes a dataset with real stereo images and a dynamic grasp simulator. The user's method should suggest grasps using the stereo images. These grasps will then be evaluated by the dynamic grasp simulator.
In the past years, different vision-based method have been proposed for grasping unknown object. However, the methods have been tested using different hardware and experimental setups. VisGraB makes it possible to compare the different methods, by providing a standard benchmark.
The video below illustrates the use of the benchmark in our own work .