source
J. Leitner et al., “The ACRV picking benchmark: A robotic shelf picking benchmark to foster reproducible research,” 2017 IEEE International Conference on Robotics and Automation (ICRA), Singapore, 2017, pp. 4705-4712.
Due
Strictly speaking, APB is not a data set, but relying on data set design contest crawl. The original intention of the proposed APB:
1, the existing environment can not be copied Robotics Competition (such as Amazon Picking Challenge (APC)), individual researchers can not rebuild a competitive environment to validate their algorithms. (The main problem)
2, champion race almost perfect setting to complete the task, leading to not get better promote research to solve the problem
3, content than a single race, such as including only visual perception, capture, detection, not a common assessment of the robot system
4, less the degree of difficulty of the task, unable to meet the research needs of researchers and large entry of God at the same time. and many more. .
To solve the above problem, the authors designed ACRV Picking Benchmark (APB) used in object-based APC competition.
objects and shelf
Cabinet (shelf) used in IKEA Kallax cabinet.
life objects used in some common object, to an object-oriented tasks have different degrees of difficulty, objects descending, light to heavy, reflective, black, transparent, and so have a total of 42.
Providing data for each data set of the object with realsense taken, and the corresponding tag.
In addition to data collection, the authors also corresponds to the design of each object placement in the cabinet, a variety of tasks, assessment methods, etc., because there with less, there is no careful study.
Dataset link: http: //juxi.net/dataset/acrv-picking-benchmark/
