Computational Science Technical Note CSTN-124

CSTN Home

MechBench: Benchmarking Motion Control of Vehicles with Mechanical Constraints

A. P. Gerdelan

Archived January 2011

Abstract

This paper describes a method for evaluating machine-learning control systems for vehicles. There is no standardised, objective framework for evaluating mechanically simulated vehicles in a physically simulated environment, where a large number of behvarioural constraints exist. This paper provides specifications for such a framework, including environment design, vehicle specifications, and evaluation criteria. Special consideration is given to measuring the adaptivity of vehicle controllers to change in environment. The framework is also used to evaluate an existing control system with results discussed. The framework has potential for comparison of control algorithms or for machine-learning in an objective self-evaluation role.

Keywords: steering behaviour; evaluation frameworks; machine learning; genetic-fuzzy systems; mechanical simulation; rigid-body dynamics.

Full Document Text: PDF version.

Citation Information: BiBTeX database for CSTN Notes.

BiBTeX reference:

@TECHREPORT{CSTN-124,
  author = {A. P. Gerdelan},
  title = {MechBench: Benchmarking Motion Control of Vehicles with Mechanical
	Constraints},
  institution = {Computer Science, Massey University},
  year = {2011},
  number = {CSTN-124},
  address = {Albany, North Shore 102-904, Auckland, New Zealand},
  month = {January},
  timestamp = {2011.05.16},
  url = {http://www.massey.ac.nz/~kahawick/cstn/124/cstn-124.pdf}
}


[ CSTN Index | CSTN BiBTeX ]