Research

Nonprehensile throwing, catching, and balancing of a disk-shaped object by a two-link planar manipulator mounted with a disk-shaped end effector is shown below. Given a goal position which is out of the robot’s reachable space, the required release position and velocity for throwing are first determined. Catching manipulation is followed by balancing control to prevent the object from falling after catching.

Among many types of nonprehensile manipulation, one type that we are particularly interested in is contact juggling in which a smooth object rolls on another smooth object. The generalized problem to solve is "Given a parameterization of the surfaces and the desired trajectory of the object in space, how should the manipulator be controlled to achieve the desired object motion?"

Nonprehensile throwing, catching, and balancing of a disk-shaped object by a two-link planar manipulator mounted with a disk-shaped end effector is shown below. Given a goal position which is out of the robot’s reachable space, the required release position and velocity for throwing are first determined. Catching manipulation is followed by balancing control to prevent the object from falling after catching.

Although autonomous robotic systems are improving rapidly, the need for teleoperation is still present in many hazardous and unknown environments. The improvement of operator performance is effected directly by the experience of the operator, the difficulty and fragility of the task and environment, and the robustness of the robot being controlled. To assist operators, various types of feedback are utilized in teleoperated systems. A pilot study on operator-customized haptic guidance functions is assessed to establish what affects it has on the operators performance and the operators comfortability and workload with a customizable teleoperated system we developed for this study.

Whole body vibration (WBV) is associated with various adverse health outcomes among professional vehicle operators. Due to rough terrain, off-road vehicle operators are likely exposed to not only a high level of WBV but also different types of WBV, i.e. non-vertical and/or rotational accelerations. To evaluate the effects of these different WBV exposures on human responses, it is necessary to correctly estimate the acceleration of the vehicle by taking account of not only translational, but also rotational motion. The main objective of this research is to develop an algorithm to accurately estimate global acceleration of a vehicle using Fourier analysis, magnitude-based filtering and inertial navigation with an inertial measurement unit (IMU). The correctly estimated global accelerations of an off-road vehicle can be used to replicate the field-measured vibration on a 6-DOF Stewart platform in a laboratory setting.

Mobile robots and manipulators have a wide range of practical applications. In those applications, it is of our interests to solve the planning and tracking control problem in an integrated, unified framework using various control theories such as differential flatness. This type of control framework would be particularly useful when there exist disturbances/uncertainties, nonholonomic constraints, and/or under-actuation with UGVs (Unmanned Ground Vehicles) and UAVs (Unmanned Aerial Vehicles).

Robot Operating System (ROS) is an open source platform over an existing operating system providing various types of robots with advanced capabilities from an operating system to low-level control, which is gaining growing popularity in the robotics community. We have developed an experimental platform controlled by a 7-DOF robotic manipulator arm run by Robot Operating System (ROS). The feedback control of balancing a ball on a plate-type end effector has been demonstrated and the pilot system is being improved to perform more complicated tasks requiring more number of joint control.

Neural-Networks (NN) provides a different way to control robotic systems compared to traditional control methods. We've worked on the development of a neural network (NN)- based feedback controller in order to compensate for the errors caused by using an approximated dynamic model in controller design. The controller consists of two subcontrollers working in parallel: base linear controller and NN-based PID compensator. It is based on a PID controller with adjustable gains and a neural network is used to update the PID gains during control process, which was demonstrated with a ball-on-plate system built for this study.