Investigating the effect of grasp stiffness on grasp success prediction
Assessing whether an acquired grasp will successfully withstand the expected disturbance load is a necessary step in autonomous robotic grasp planning and execution pipelines. To this end, features obtained from tactile sensors on the hand's fingerpads have shown to be highly valuable [1, 2].
The aim of this thesis is to extend the recent grasp success prediction framework in [1] with an explicit notion of the grasp stiffness. To this end, the student will set up an experimental environment using ABB's YuMi robot and perform a series of grasping experiments to investigate the effects of grasp stiffness. An optional extension of the thesis is to augment the analytic framework with a model learned from data collected during the grasping trials. Overall, this thesis will have a strong scientific and experimental character.
Tasks:
- Augment an existing grasp success prediction metric with a notion of stiffness
- Implement a grasp stiffness controller on the ABB YuMi platform available in the lab
- Perform a series of grasp experiments to evaluate the new metric
- Use machine learning to incorporate the collected experimental data into the grasp success prediction approach
Necessary skills:
- Solid C++/Matlab knowledge
- Experience in the ROS ecosystem would be helpful
Contact: Robert Krug
[1] Krug, Robert, et al. "Analytic grasp success prediction with tactile feedback." 2016 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2016.
[2] Bekiroglu, Yasemin, et al. "Assessing grasp stability based on learning and haptic data." IEEE Transactions on Robotics 27.3 (2011): 616-629.