My current research focus on the confluence of robotics and sensor networks. While many researchers study different domains of sensor networks, my research will focus on incorporating mobility in sensor networks. Sensor networks can benefit from mobility in many different ways. Mobile robots can be used to deploy sensors over a terrain, to do on-site calibration to reduce sensor drifts, and to provide power supply to a sensor when its battery runs low. When mobile robots are mounted with sensors, they can do active sampling and significantly reduce the number of sensors deployed in a large terrain. Furthermore, in some applications like plume tracking and neutralizing, a team of fully mobile robots with sensors and actuators can be exploited. These topics have attracted some attentions in the past few years, but the research in this domain is rather nascent, and many novel interesting problems arise. We are interested in problems such as how we should adopt the sensing and communication information to navigate and coordinate robots in networked sensors environments. For instance, how to navigate a robot through a field of networked sensors where global information such as maps and localization is not available. More generally, by our research, hopefully we can have better understanding on the question that how we should use robot mobility to achieve maximum utilization of a sensor network system.
The research in my master’s program was mainly concerned with the path planning problem with general end-effector constraints for robot manipulators with many degrees of freedom, in environments with obstacles. For example, a robot manipulator holding a glass of water should keep the glass vertically up all the time, which imposes a constraint on end-effector orientation; or, in some applications, the end-effector may be constrained to move in a plane, which imposes a constraint on end-effector position. In our problem scenario, the robot must satisfy the end-effector motion constraints while avoiding obstacles.
Most of the previous research used local Jacobian-based control techniques, which may fail to find the solution even in simple environment, say with two or three obstacles. We proposed two global path planning based approaches to deal with the problem. The first approach is adapted from the existing Randomized Gradient Descent (RGD) method for closed chain robots. Intuitively, with constraints on the end-effector, the problem can be treated as a closed-chain robot path planning problem. As the most path planners, this approach works in robot parameter space, the configuration space. The second approach is a radically different planning algorithm, called Alternate Task-space And C-space Exploration (ATACE). This approach works in both the workspace and the configuration space, and it use workspace heuristic to effectively and efficiently guide the configuration space search.
The motion planning kernel (MPK) is a software system designed to facilitate development, testing, and comparison of robotic and geometric reasoning algorithms. Examples of such algorithms include automatic path planning, grasping, etc. The system has been designed to be open and extensible, so that new methods can be easily added and compared on the same platform. More more information about MPK, please click here.