Modeling, Estimation, and Control for Navigation of Flexible Continuum Robots
Continuum manipulators are robots with long, slender, and flexible bodies that can be deformed into smooth three-dimensional curves. Their snake-like shape and intrinsic compliance make continuum manipulators useful for applications that require movement through unstructured and constrained environments. Examples include search-and-rescue, confined space inspection, and minimally invasive surgery. This dissertation addresses navigation of a flexible continuum robot to a target location. Because of the body's flexibility, accurate mechanics-based models of continuum manipulator motion are sensitive to physical parameters that must be experimentally determined and may change with time. Further, contact with the environment which is often unknown, can significantly affect the accuracy of these models. Because of these difficulties, this thesis focuses on simple, empirical models that are coupled with conservative control methods that are robust to model inaccuracy. These techniques are demonstrated on three different continuum robots: a minimally invasive surgical instrument known as a steerable needle, and two types of soft continuum robots that grow from the tip.
Feedback of robot configuration is needed for control of a continuum manipulator. In applications such as search-and-rescue and surgery, it may only be possible to directly measure a part of the robot's backbone rather than its entire length. To overcome this difficulty, we develop an estimation method to infer the state of the manipulator's backbone by combing multiple partial observations of the moving robot in time. We demonstrate the method using freehand ultrasound and steerable needles, which is part of a larger effort to use robotic needle steering for percutaneous ablation.
Also described is a new soft continuum robot that has a novel extension (``growth") degree-of-freedom that is realized using pneumatic eversion, in which new material is fed through the body and everted at the tip. The growth degree-of-freedom enables movement of the robot in a direction that is always tangent to its backbone curve. Among other benefits, it simplifies control of the robot by decoupling steering and movement. We introduce a new soft artificial muscle to steer the robot, model the combined growth and steering of the system, and demonstrate autonomous navigation to targets using camera feedback.
Finally, this thesis introduces a new kinematic model that describes the motion of a soft continuum robot that is in contact with obstacles. Validation experiments show that the model can correctly predict the configuration of a robot that grows through and interacts with a cluttered environment. Further, we show that obstacle interaction can be beneficial for navigation of the robot because obstacles can passively steer the robot and reduce uncertainty in its motion. A motion planner for the soft continuum robot was built on this model. Rather than strictly avoiding obstacles in the environment, the planner may choose to exploit obstacles when useful for reaching a destination. Validation experiments show that modeling obstacle interactions in the planning stage can increase the likelihood of successfully navigating to a destination.