The primary purpose of the project is to develop methods that dramatically increase the utility and reliability of a robot's sense of touch. The PIs' proposed study of robot haptics (perception and manipulation through active touch interaction) particularly addresses the use of robots in unstructured situations where little may be known about the objects to be manipulated and where visual information alone is not sufficient to guarantee robust task performance.
This research will consist of three main components: 1. Probabilistic sensor processing methods for haptics. The strength of probabilistic techniques is their ability to fuse together many streams of noisy sensor data (such as contact information) based on a given physical model. 2. New active perceptual methods. As with human senses, active exploration is required to provide sensors with a rich source of information about physical properties of the world. 3. Autonomous strategy selection. Often it is necessary to switch between manipulation and exploration in order to accomplish a task. A key contribution of this research will be a decision-theoretic probabilistic method for determining when manipulation is likely to succeed and when further haptic exploration is required.
In order to support the proposed robot haptics research, the construction of the Probabilistic Manipulation Experiment Table (PMET) is proposed. The PMET will consist of one or two small manipulating arms with instrumented 'fingertip' end effecters that support force sensors, accelerometers, and tactile sensors. An interactive control software package, the Probabilistic Robotics Studio (PRS), will be developed to provide a graphical interface to configure experiments and view probabilistic models in real-time. Initial phases of research will develop probabilistic models and exploration strategies that characterize properties of contact (including contact slip) and non-visual properties of objects (frictional coefficients, center of mass, etc.).
Intermediate phases will demonstrate autonomous strategy selection during robust one-finger manipulation of objects in the plane. The final phases will include two-finger exploration, two finger manipulation, and lifting objects.
Increased understanding of touch sensory information from this work will give the robot design community better construction techniques for robot hands and their sensor systems. The PMET and PRS will provide an interactive experimental platform that can be used to train robotics students and explore new ideas. The PIs plan summer internships for high school students in the area around Stanford University to allow experimentation with the PRS. In addition, the PIs will aid the interns in development of presentations for K-12 students.
