Chapters 5 and 6 illustrate that the BOXES methodology works very well in prolonging the duration of motion of a freely hinged pole that is attached to a constantly moving cart. Chapter 5 considered simulations of the system that ignored the effects of friction in the hinge, while Chap. 6 illustrated the attachment of a physical pole and cart system to the algorithm using custom firmware. This chapter focuses on how to ensure that the training sequence precludes any preferential treatment of any region in the control space that might be inadvertently produced. To ensure this, auto start systems are described that control how either system is parked after a failure, and subsequently restarted in a random but valid state space region. During the restart process, the motor direction on the trolley does not necessarily match the corresponding value in the signature table. It is imperative that the system be not over-trained in any one trajectory, such as is the case if, for example, the pole is held vertical in the center of the track and released. Two strategies, one for simulations and one for the physical system are described. In simulation studies it is relatively simple to ensure that initial values of the state variables are truly random from run to run and distributed across the vector space. In the physical system, the autostart firmware was designed to promote random starts of the system and also allows a manual override of the control logic using LEFT/RIGHT push button commands.
- 1.Boyes J. (1977) MPhil (CNAA) Thesis. Liverpool Polytechnic.Google Scholar
- 2.Grant, A.S. (1990) Modelling Cognitive Aspects of Complex Control Tasks. PhD Thesis Section 3.2 University of Strathclyde Department of Computer Science http://www.simongrant.org/pubs/thesis/3/2.html (Accessed April 1, 2011)