Multi-Objective Control of Robots
Dimitar Dimitrov, Pierre-Brice Wieber, Adrien Escande · Journal of the Robotics Society of Japan · 2014
Prior to linear programming it was not practical to explicitly state general goals and so objectives were often confused with the ground rules for the solution.(...) Thus the means to attain the objective becomes the objective in itself, which in turn spawns new ground rules as to how to go about attaining the means (...).These means in turn become confused with goals."G. B. DantzigEven though over 20 years old, the above quote still reflects well a common practice in the field of robotics.That is, not establishing a clear separation between: (i) what one wants to achieve, and (ii) how this must be done.Using high-level goals for posing real-world problems in mathematical terms can be advantageous since, at the level of modeling, one need not consider the particular technique for approaching their solution.In fact, being able to abstract the problem formulation can be viewed as a part of the revolutionary development that followed the birth of the field of linear programming [1], because practitioners could be trained to cast situations (of potentially great complexity) in terms of a set of general goals and rely on available solvers developed by specialists in the fields of numerical analysis and optimization.Such separation leads to a reliable solution process.In this note we argue that the ability to define typical robotic problems in terms of lucid goals is beneficial not only from the point of view of clarity of presentation, but also for efficiency of computations.Our main point is that this can be achieved through the explicit use of multi-objective formulations.By means of several 原稿受付 2014