Throughout the project, the SARAFun team needed to manage a large number of data-sets, generated and collected by various means, i.e. sensors, cameras, robots and direct interactions with users (e.g. interviews and questionnaires). By the end of the project, 9 different data-sets have been produced through the SARAFun’s technical activities, with almost all the partners being data owners and/or producers. The data-sets are the following:
- CERTH.FAIM2017Dataset : Data-set used for keyframe extraction in laboratory environment. An instructor person will pick up two small objects and, afterwards will assembly them. This data-set was used in “Teaching Assembly by Demonstration using Advanced Human Robot Interaction and a Knowledge Integration Framework”.
- CERTH.IJERTCS2017Dataset : Data-set of responses from users that had to rate the SARAFun HRI interface. The data-set has been used in the paper with title “An Advanced Human-Robot Interaction Interface for Collaborative Robotic Assembly Tasks”
- CERTH.CVPR2016Dataset : Data-set of RGB and depth images reflecting two usage scenarios, one representing domestic environments and the other a bin-picking scenario found in industrial settings.
- CERTH.SnapFitForceProfiles : Data-set is used for training and testing a machine learning classifier in order to achieve real-time detection of successful snap-fit assemblies. The data-set contains force profiles on the axis of motion (assembly), captured during a robotic and a human assembly process of two different snap-fit assembly types, namely cantilever and annular.
- CERTH.ContactEvaluationData : Data-set generated by logging wrench forces of the robot’s F/T sensor in various contact configurations between the assembly parts. The data-set contains ROS bag files of the logged forces.
- ULUND.TransientDetection : Data-set used for evaluation of a recurrent neural network (RNN) for recognition of transients, in order to detect events during robotic assembly. Inputs are robot joint torque data. Outputs are probabilities that the event is occurring, as estimated by the RNN.
- UNIBI.TactileData : Tactile data for slip detection experiments. Various objects are hold by two KuKa robots between two tactile sensors with different initial forces and released to create slippage events.
- ABB.ExperimentalVerification_GraspQuality : Data-set used for measuring grasp quality of automatically design fingers for industrial robots. The data-set is available in Microsoft Excel format (i.e. .csv).
- TECNALIA.Human_Performance_of_Bimanual_Assembly : Recording of experiments in which volunteer human subjects performed a sliding insertion task using instrumented objects to measure the kinematics and interaction forces during uni-manual and bi-manual manipulation.