Learning Agents: Learning Automata and Need-based Drive Reduction

Learning Agents: Learning Automata and Need-based Drive Reduction

A Multi-agent System composed of independent Learning Agents motivated by Abraham Maslow’s Hierarchy of Needs theory

Figure below: Seen from overhead, this is a screenshot of the 2D maze created for a Multi-agent System composed of independent Learning Agents. At program start, the icon hosts (affectionately nicknamed BUGS) were given random placement. The icon BUGS were autonomous and mobile, and were endowed with multiple independent Learning Automata. In this FIGURE, the BUGS have not yet completely taken their places to the sides of the baffles in the simulated maze. However, once the simulation began, the BUGS would quickly fall to the floor of the maze and start moving. The four gray circles on the floor of the maze were feed-stations for the prey BUGS. Predatory BUGS (blue dots) and prey BUGS (red dots), respectively, emitted unique simulated scents of controllable intensity. These odors provided rich, simulated olfactory stimuli to the entire company of BUGS in the maze. Abraham Maslow’s Hierarchy of Needs theory provided the motivating framework for the reducible drives embedded behind the Learning Automata. During run-time, the icons were free to move about the maze until they were eaten, their energy levels were depleted, or they adapted, found available food, avoided their predators, and survived.

What is the model (in Abstract)?

The Figure above is a screenshot from a Multi-agent System (MAS) where foraging behavior, survival tactics, and inter-agent socializing in vehicular icon BUGS were steered by a novel, need-based, drive-reducing Learning Automata architecture.  Each “agency” in this MAS was composed of four, independent, Turing P-Type, unorganized machines acting as individual “agents.”  At load-time, the user could select the initial number of predatory and prey host icons for instantiation within an enclosed maze.  Within each respective agency (a BUGS icon), one of four LA would be selected by afferent conditions to equilibrate over sensory affect provided by simulated host icon olfaction and an isopraxis ethology.  All behaviors in the hosts were the emergent consequence of self-organization.  Appearing as icon hosts with differences in color, size, and markings, predators were distinguishable from prey.  The program used two physically separate computing platforms in distributed fashion: a maze computer and a control computer.  The former machine generated the simulation of a multi-baffle, porous-wall (open trellis-like) maze containing the icon hosts.  The latter machine generated efferent control signals for the predatory and prey hosts, and received sensory afferents as inputs from the former machine in return.  Olfaction, a haptic sense, and hunger/satiation were the host afferents.  Motor commands were the efferents.

Who is behind this model?

This model was developed by Ovi Chris Rouly from Department of Computational Social Science, at George Mason University as part of regular course work associated with the PhD program in Computational Social Science.

Can I read about the model?

Yes. A short paper was written that describes the model and its operation. That paper was accepted to the 8th International Conference on Intelligent Technologies, Sydney, Australia for delivery at the conference and for publication in their proceedings volume. You can download the paper here.

Rouly, O. C. (2007, December). Learning automata and need-based drive reduction. In Ha, Q. P., & Kwok, N. M. (Eds.). Proceedings of the 8th International Conference on Intelligent Technologies, Sydney, Australia.

Click here to see more models