Core Concepts
Guided motion planning formalizes the intuitive notion of guided search by defining the concept of a guiding space, which encapsulates many seemingly distinct prior methods under the same framework. It provides a new information-theoretic method to evaluate the quality of guidance, which matches intuition when tested on known algorithms in various environments.
Abstract
The paper presents a framework for guided motion planning, which formalizes the intuitive notion of guided search by defining the concept of a guiding space. This new language encapsulates many seemingly distinct prior methods under the same framework, allowing the authors to reason about guidance, a previously obscured core contribution of different algorithms.
The authors suggest an information-theoretic method to evaluate guidance, which experimentally matches intuition when tested on known algorithms in a variety of environments. The language and evaluation of guidance suggests improvements to existing methods and allows for simple hybrid algorithms that combine guidance from multiple sources.
The paper first introduces the concept of a guiding space, which is an auxiliary space that helps estimate the value of exploring each configuration in the configuration space. The authors define a guided search algorithm that uses the guiding space to inform exploration. They then discuss three main categories of guiding space methods: robot modification, environment modification, and experience-based guidance.
Next, the authors propose a new metric called sampling efficiency to evaluate the quality of guidance. This metric measures the difference between the distribution of samples produced by the guiding space and a target distribution derived from an optimal sampling distribution with oracle access. The authors test this metric on various algorithms and environments, showing that it captures properties of algorithms otherwise obscured by traditional metrics such as runtime or number of samples.
Finally, the authors refactor existing algorithms to fit the guided motion planning framework and demonstrate the benefits of this approach, including the ability to create hybrid algorithms that combine guidance from multiple sources. The experiments show that the hybrid approach can outperform its individual parts, highlighting the value of framing sampling-based motion planning as guided search.