I want to build intelligent robots -- robots that would perceive and manipulate physical objects as cleverly and effectively as humans. At present, robots don't understand the real world very well, so we have to program them in detail. As a result, robots are clumsy, virtually blind, and do not react well to unexpected events. A better robot would have an internal model of the physical world, allowing the robot to predict the effect of its actions, or to plan a sequence of actions to achieve its goals.
The chief complication is uncertainty. The real world does not always behave as expected. The robot's motions are imprecise, the sensors are noisy, and all of the relevant physical parameters, such as the coefficient of friction, are hard to predict. A robot can deal with uncertainty in two ways: it can adjust its model and its actions based on sensory feedback, and it can choose actions that are insensitive to variations in the world. Underlying both of these options, a robot must include uncertainty in its model of objects and physical processes.
My research efforts are organized around the following topics:
Mechanics of manipulation: To build robots that model the physical processes of manipulation, we first must understand these processes ourselves. Manipulation takes place in a world often dominated by friction and collisions.
Robotic origami: Origami, the Japanese art of paper-folding, represents the pinnacle of human manipulation, and is a challenging task domain for robots. There are great problems to address: modeling and simulation of deformable materials, kinematics of folded paper, and planning and control of a variety of interesting manipulation modes.
Manipulation without hands: Many mobile robots are mobile manipulators, even though they do not have hands. In soccer, for example, the goal is to manipulate (or perhaps we should say pedipulate) the ball. Rather than assigning the manipulation task to a manipulation subsystem (an arm) we ask what the mobile robot is able to do with its entire set of resources. This view is scientifically more interesting and leads to more elegant and economical systems.
Locomotion without legs: In a similar vein, by exploiting all available resources, a mobile robot can recover from error conditions where its legs or wheels are unable to do the job. For example a high-centered robot might induce a rocking motion by using its legs as reaction mass, and recover from its predicament in much the same way a human would.
Shortest paths for mobile robots: One of the most fundamental characteristics of a mobile robot is its shortest paths. The shortest paths are useful in motion planning and control. To find the shortest paths we assume a velocity bound and find the time optimal paths.
|Research Interest Keywords|
|artificial intelligence, assembly, control, factory and warehouse automation, manipulation, manufacturing, mechatronics, mobile robots, motion planning, obstacle avoidance, planning|
|The Robotics Institute is part of the School of Computer Science, Carnegie Mellon University.|
Contact Us | Update Instructions