Back
to indexD. Langer and J. K. Rosenblatt and M. Herbert
Summary
Wow. A great article. It completely outlines their algorithms in detail, breaking the system down into three components. 1) perception module to identify untraversable regions of terrain, 2) local map management module 3) planning module to issue commands to the vehicle controller. The vehicle proposed is able to deal with rugged terrian, potholes, and can persist over long periods of time. Behavior-based programming is used as the decision algorithm, which is the culmination of many years of research at CMU combining distributed efficiency with independant goal finding, avoiding sensor fusion, and is very robust. However, it is sensitive to noise since each module isolates the processing of its sensor data. We may avoid this approach by a little sensor fusion of related sensors.
Methods
1. Classify the terrain into traversable and untraversable regions based on range finding laser data. Untraversable regions are identified by i) height variation within a cell, ii) orientation of the vector normal and iii) presence of a discontinuity of elevation in a cell. An algorithm for determining ideal grid resolution is defined.
2. Manage the map locally. Throw out data not in the active region. The local map has just the traversable/untraversable values stored relative to the vehicle. Map update is fast since the region is small. The vehicle may only traverse what is in the current field of view, so sharp turns are discouraged by introducing artifical obstacles just outside the FOV.
3. Plan the path using behavior programming. Construct an arbiter which analyzes the inputs from the different behaviors, based on a +/-1 counting regime and a voting scheme. Again, the algorithms are detailed - see the paper! Behaviors include the obstacle avoider, the goal seeker, the drive straight behavior, and the maintain turn behavior.
Keywords
obstacle detection, map management, arbiter, behavior programming, traversable regions, ideal grid resolution formula
Rating
10
Bibtex Entry
@article{ langer94behaviorbased,
author = "D. Langer and J. K. Rosenblatt and M. Herbert",
title = "A Behavior-Based System For Off-Road Navigation",
journal = "IEEE Journal of Robotics and Automation",
volume = "10",
number = "6",
pages = "776--782",
year = "1994",
url = "citeseer.nj.nec.com/langer94behaviorbased.html"
}