A robotic hand reaches into a bag and gropes about, and after a moment it emerges with a set of keys. While holding the ring firmly with two fingers, the others are used to separate and grip one key in particular. The robot then begins to insert the key into a lock. The lock is old, and the key worn, and the hand feels the key begin to slip in its grasp. After tightening the grip and changing the key's angle in the lock, a satisfying sound and tell-tale vibration indicates the key can be turned and the door opened.
Our projects in mobile grasping and manipulation rely on biologically inspired solutions to hand design and sensor integration and interpretation challenges. Robotic hands need to be strong and lightweight, and allow for variable compliance tailored to the task at hand. At the same time "mechanical intelligence" which allows the hand to passively conform to objects using a small set of actively controlled grasp types relieves a large portion of the control burden. Tactile sensing suites on hands and arms, in contrast to physically separated vision systems, are still relatively uncommon, fragile, and limited in scope and coverage. As we improve the sensors themselves and cover more of our robots' hands and arms, we need to explore how to effectively use that information to improve grasping and manipulation.
A number of advances will be needed to realize the goals of robust grasping and manipulation on mobile robotic systems. Some of our current work towards those ends is outlined below
- Tactile sensors and artificial skins are being designed with sensing capabilities inspired by that of humans. Low frequency normal and shear pressure sensing are being combined with high frequency dynamic tactile sensing to produce a single skin which can detect light touch, high pressure, and slip.
- Under-actuated compliant hands are being developed for a couple of applications, including underwater exploration (Red Sea Exploratorium) and mobile manipulation (DARPA Arm-H). Common themes include the ability to selectively lock or release joints and to vary the stiffness of joints to increase the range of poses that the hands can assume.
- Arm-H (private page) A selectively compliant, under-actuated hand is being designed in collaboration with SRI and Meka Robotics. The selectively compliant mechanism should allow the hand to be quite dexterous despite being under-actuated, and allows for simple control of how the hand passively adapts to and grasps objects. The latest in our work on tactile sensing is being incorporated to give the hand the ability to feel what it interacts with and respond accordingly.
- Tactile event sensing for grasping and manipulation. We have developed a suite of normal pressure and dynamic tactile sensors for an underactuated hand mounted on a typical industrial robot arm. This setup can identify and characterize tactile events, such as contact and slip. A particular goal is to distinguish between hand/object and object/world slippage. One application of this work is for a Robot Drilling system.
- Underwater hands and sensing -- We are developing and hand and sensors for the Red Sea Exploratorium project, in collaboration with the C.S. Dept.
D. M. Aukes and M. R. Cutkosky, “Simulation-Based Tools For Evaluating Underactuated Hand Designs,” in 2013 IEEE International Conference on Robotics and Automation, 2013. (preprint)
D. M. Aukes, M. R. Cutkosky, S. Kim, J. Ulmen, P. Garcia, H. Stuart, and A. Edsinger, “Design and Testing of a Selectively Compliant Underactuated Hand,” 2013. (In Review)
Aukes, D., Kim, S., Garcia, P., Edsinger, A., & Cutkosky, M. R. (2012). Selectively Compliant Underactuated Hand for Mobile Manipulation. 2012 IEEE International Conference on Robotics and Automation.
Aukes, D.A., Heyneman, B., Duchaine, V., and Cutkosky, M.R. "Varying spring preloads to select grasp strategies in an adaptive hand" IEEE IROS 2011, San Francisco, Sept. 2011. (slides)