Localization in Multi-Robot Settings

From Thinclabwiki

Revision as of 20:51, 6 April 2012 by Anousha 1986 (Talk | contribs)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to: navigation, search
What is Localization in Multi-Robot Settings?
The Turtle-Bot robotic platform.

In a multi-agent environment, agents share the resources and make decisions based on the knowledge representation of the resources. In order for the agents to make rational decisions, they must be aware of other agents' actions either based on the observations that they make or the behavioral model they have for other agents. In the multiagent settings, the agents in the environment could be cooperative, meaning that they communicate in order to have a unique knowledge representation and achieve a unique goal, or they could be independent with different goals and no means of communication.

Application domains such as localization and mapping exhibit characteristics such as multiple cooperative or independent robots moving in the environment, exploring and trying to localize and map simultaneously. In the environment that is shared by multiple independent robots, such as search and rescue in disaster areas, the robots explore the environment and move obstacles for the search purposes, that can substantially affect other agents’ model representation of the environment (maps) and therefore the rational decision made by the other robots that are not aware of the new changes.

Project Description
Initial configuration of the environment. The blue object on top of the figure is robot i and the blue object on right bottom corner is robot j.
Initial representation of the particles for i.

In this project, we use nested particle filtering approach to track the subject robot (robot i) and other independent robot (robot j) in a dynamic environment. We generalize particle filtering to multi robot settings in order to localize the subject robot in a partially observable environment using landmarks. This often entails simultaneously tracking the uncertain location of the other robot(s). We assume that the exact locations of the landmarks are known to the subject robot. Our focus is on the subject robot's localization at its own level in the presence of others who may not be cooperative.

We introduce a nested set of particles to track the subject robot and others, and recursively project these particles as the subject robot moves and makes observations. In nested particle filtering, for each of i’s hypothesized pose (particle), robot i maintains a set of hypothesized poses for j. We adopt a subjective view from the perspective of i. Hence, each particle of i contains i's pose and a set of particles reflecting j's possible poses.

We extend Rosencratz et al.'s laser tag domain for our experimentation. In particular, we generalize the problem by assuming that the subject robot is itself not localized. We adopt the perspective of a robot i whose task is to tag another robot j and then proceed to reach j's base within a certain amount of time steps. The physical environment shared by both robots is populated by multiple landmarks of different colors and sizes. Two of these objects are distinguished and serve as bases of each robot. If robot j is tagged, i then uses path planning to move toward j's base while j proceeds towards the nearest object with the aim of displacing it thereby possibly disturbing i's localization and slowing its progress toward j's base.

We simulate the laser tag domain and perform our experiments in a 3D environment using Microsoft's Robotics Developer Studio.

  • Prashant Doshi, (PI)
  • Anousha Mesbah, MS. student, CS
  • Kenneth Bogert, Ph.D. student, CS
  • Eliot Beckham, Young Dawg, High school senior, 2011
  • Harrison Katz, Young Dawg, High school senior, 2011
  • Joshua Brown, Young Dawg, High school senior 2010
Personal tools