Computation Reuse

Given two environments sharing similar obstacles, previous methods treat them as two completely different problems. Our study shows that by carefully storing and reusing the computation for one pronblem, we can solve the other efficiently.

image Collision Prediction: Conservative Advancement Among Obstacles with Unknown Motion
papers: IROS 2014 | IDETC/CIE 2014 | WAFR 2014 |
In this work, we assume an obstacle moves along some unknown trajectory with bounded velocities. Instead of replanning periodically at fixed time interval, we propose to make a conservative and tight estimation of the time that robot remains safe on its current path and replan only when necessary.
image Finding Critical Changes in Dynamic Configuration Spaces
papers: IROS 2011
In a dynamic environment, this planner is able to identify all topological changes of free configuration space. As a result, it provides a more complete representation of configuration space. Besides, it reuses computation and avoids unnecessary updates so that efficiency is greatly improved.
image Planning Motion in Similar Environments
papers: RSS 2009
We developed a new method called RU-PRM which reuses the computation for previously solved problems. RU-PRM stores the local roadmap built around each C-obstacle. When a new environment is given, it matches the C-obstacles and loads the matched roadmaps.
image Hybrid Motion Planning Using Minkowski Sums
papers: RSS 2008
This new motion planner called M-sum planner applies the idea of Minkowski sum to generate configurations uniformly on the surface of the C-obstacle.

List of MASC Research Pages
Computer Science @ George Mason University