Discipline: Technology and Engineering
Subcategory: Civil/Mechanical/Manufacturing Engineering
Kenneth Gutierrez - University of California, Los Angeles
Co-Author(s): Veronica J. Santos, University of California, Los Angeles, CA
Many different methods have been created to enable robotic manipulators to perform everyday human tasks, such as unloading a dishwasher. Computer vision has been used extensively to perform these tasks but performance and robustness can be greatly improved. One advantage that human fingertips have over current dexterous manipulators is the use of haptics (sense of touch) to relay information about the grasped object and the local environment. The use of tactile sensors has proven beneficial for object manipulation since information about the contact is available to the system. By incorporating ‘haptic artificial intelligence’ into robotics and prosthetics, haptic-reliant tasks (e.g. object handovers, tool manipulation) can be performed more efficiently by current robotic manipulators.
The use of multimodal fingertip sensors will allow the manipulator to predict the motion of a grasped object based on the fluid pressure, vibration, and skin deformation. A Barrett Hand (Barrett Technology) equipped with BioTac (Syntouch LLC) tactile sensors is used to grasp a parallel-faced object with two digits. The Barrett WAM arm is used to perturb the grasped object in random directions ([0:10:360 deg]). In addition, we varied the grip force (LOW, HIGH) and perturbation velocity (2, 4 cm/s). By analyzing the DC Pressure, AC Pressure, and electrode impedance output from the tactile sensor, the direction of the object motion can be perceived. Preliminary results suggest that changes in fluid pressure are associated with distal-proximal motion while skin deformation is associated with radial-ulnar motion. Next steps are to create offline Support Vector Machine classifier models and online Hidden Markov Models for perception of movement direction and decision making.
Not SubmittedFunder Acknowledgement(s): This material is based upon work supported by NSF GFRP Grant No. DGE-1144087, National Science Foundation, Award #0954254/ 1461547.
Faculty Advisor: Veronica J. Santos, vjsantos@ucla.edu