Adaptive Path Planning in a Dynamic Environment using a Receding Horizon Probabilistic Roadmap: Experimental Demonstration

If dynamics and constraints are nonlinear, there is no closed-form analytical solution

Gradient and potential field methods can be useful in nonlinear domains, with restricted observability, but are strongly influenced by initial search location, as nonlinear constraints can introduce many maxima/minima

Gradient methods may require special differentiation because of non-smooth constraints

The main idea of the paper is to use PRMs to build a global connectivity graph, which may need to have edges removed as the agent only senses obstacles nearby, so replanning is still needed as locations that were thought to be free turn out to be closed

Planning is done purely in state space, then motion primitives stitch the states together.

They plan in a quadratic cost function

The use a helicopter domain that they provide simple dynamics equations for

The motion primitives used for planning ensure the helicopter does not enter a bad state

They also tested the approach on an actual helicopter