Motion Planning Under Uncertainty
Ali–akbar Agha–mohammadi, Sandip Kumar, Suman Chakravorty · 2012
A basic problem that has to be faced and solved by autonomous vehicles, on which this chapter will focus, is the problem of motion planning. It involves generation and execution of a plan for moving around an environment towards a designated goal, or to accomplish a desired task, while avoiding collisions with obstacles in the environment. Moreover, it is desirable to optimally use the available resources to achieve the goal, thereby minimizing some “cost” function. There are many established techniques to solve motion planning problem in a deterministic framework ranging from optimal control method [1], grid world approaches [2] to randomized sampling-based motion planners [2–4]. However, real-world systems are not deterministic, and their evolution involves uncertainty due to surrounding environments or internal parameter variance. Hence, in the real world, these deterministic algorithms are applied along with some trajectory tracking techniques, which accounts for the uncertainty in the system [5]. Another approach to solving the motion-planning problem is to take uncertainty into account while solving the problem. Introduction of uncertainty in the motion-planning problem increases the complexity of the problem. Uncertainty in the system can be caused by two scenarios, one due to sensing uncertainty and the other due to process uncertainty. Sensing uncertainty arises from sensor noise during measurements or an uncertain environment, that is, partial knowledge of obstacle locations in the given environment. The process uncertainty is the motion uncertainty due to presence of stochastic forcing in the system dynamics and controls.