Probabilistic Action Formulisms with Applications to Robotics
In many robot applications, an important step before deploying a robot program is to verify whether the program satisfies certain properties. There are related works consider the verification problem of robot programs under full-observations. Yet, in practice, the environment is almost never full-observable to the robot, i.e. partial observable. For instance, robot sensors are subjects to noise, physical actions are subject to uncertainty .
Before considering verification, one must design a formulation to model partial-observation where noise sensor, stochastic actions and imprecise knowledge should be incorporated. Perhaps, the most successful work in doing this is BHL’s model which combine probability theorem and situation calcula by introducing beliefs to describe the robot’s epistemic state. Yet, it fails to modeling incomplete knowledge, which is exactly how humans interact with environment.
My work focus on modeling robot’s beliefs (particularly, incomplete beliefs ) and reasoning in dynamic domain. Afterward, I will explore belief programing and planning, and verifying. So far, some successes are made in the formulation and reasoning mechanism, i.e. regression.