How important is tactile sensing for manipulation? Many of our projects are driven by the assumption that the answer to this question is "critically important". The human tactile system is superb in its ability to collect and centralize large amounts of multi-modal data. In robotics, decades of progress have provided us with numerous ways of transducing contact force and displacement to an electric signal; still, we are nowhere near the ability to build a robot hand with an integrated tactile system that would even approach its human counterpart. Our goal is to build tactile sensing systems that:
- provide very rich data on many aspects of touch;
- allow complete coverage of complex geometry and articulated kinematic chains;
- do not present unsurmountable integration challenges when built into complete robot hand.
A significant part of our work on tactile sensing is in collaboration with the Columbia CLUE Lab.
Project highlights (in chronological order):
We demonstrated a sensor that can achieve sub-millimeter contact localization accuracy over a 160mm2 area using only 4 wires. The key idea is to collect and use overlapping piezoresistive signals from multiple pairs of electrodes. [Publication: IROS 2016][Video below: piezo-resisistive tactile sensor based on spatially overlapping signals]
Continuing on the idea of using spatially overlapping signals, we moved to optics as a fundamental transducing mechanism, simplifying construction, increasing sensitivity and improving performance. [Publications: journal preprint, ICRA 2017, IROS 2016 Workshop, Humanoids 2016 Workshop][Video below: optics-based tactile sensor]
We also investigated data-driven super-resolution methods using MEMS strain gauges (Takktile by Right Hand Robotics Inc.) embedded into a hemispherical tactile dome. Our results showed accurate contact localization over a 1,400 mm2 curved surface using using just 5 individual sensors. [Publication: RA-L 2018][Video below: summary of our paper]
We then used out optical tactile sensing method to build a complete robotic fingertip. We believe this is the first finger to achieve sub-millimeter accuracy in touch localization over a complex, multi-curved three-dimensional surface. The finger is also easy to integrate on robot hands, with a single 14-wire FCC connector. [Publication: IEEE T-Mech 2020][Video below: overview of our finger]