Distinction between types of motivations: Emergent behavior with a neural, model-based reinforcement learning system

Shirinov, Elshad; Butz, Martin V.

doi:10.1109/alife.2009.4937696

Cited by 3 publications

(6 citation statements)

References 15 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Thus, the impact of each motivation needs to be adaptive. A first step in this direction has been done by adding homeostatic reservoirs, which control motivation weights [5,3]. Further work utilizing other behavioral modifiers, such as curiosity mechanisms [6], has the potential of producing even more autonomous behavior, as illustrated elsewhere for the case of a simulated mobile robot [3].…”

Section: Resultsmentioning

confidence: 99%

“…Elsewhere [3] it was shown that location-based and property-based motivations may be handled differently in that property-based motivations should be included in the propagation mechanism (value iteration) of the location-based motivations. Thus, the propagated activation level of a given neuron is a combination of goal-based activation and fear impact on the propagation:…”

Section: Setting the Level Of Fearmentioning

confidence: 99%

“…Elsewhere [2], the goals were gathered into a motivational module for a robotic system. The principles of these works were further extended in [3], where motivations were added to a neural reinforcement learning system to obtain behaviors resembling that of rodents.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

The Scared Robot: Motivations in a Simulated Robot Arm

Butz

Pedersen

2009

KI 2009: Advances in Artificial Intelligence

Self Cite

View full text Add to dashboard Cite

Abstract. This paper investigates potential effects of a motivational module on a robotic arm, which is controlled based on the biologicalinspired SURE REACH system. The motivational module implements two conflicting drives: a goal-location drive and a characteristic-based drive. We investigate the interactions and scaling of these partially competing drives and show how they can be properly integrated into the SURE REACH system. The aim of this paper is two-fold. From a biological perspective, this paper studies how motivation-like mechanisms may be involved in behavioral decision making and control. From an engineering perspective, the paper strives for the generation of integrated, self-motivated, live-like artificial creatures, which can generate self-induced, goal-oriented behaviors while safely and smartly interacting with humans.

show abstract

Section: Resultsmentioning

confidence: 99%

Section: Setting the Level Of Fearmentioning

confidence: 99%

See 1 more Smart Citation

The Scared Robot: Motivations in a Simulated Robot Arm

Butz

Pedersen

2009

KI 2009: Advances in Artificial Intelligence

Self Cite

View full text Add to dashboard Cite

show abstract

“…To combine the hunger and fear drives effectively, a dependent propagation method is implemented (Shirinov & Butz, 2009). The rule for the hunger activity propagation in a node i is given by:…”

Section: Interactionmentioning

confidence: 99%

“…Previously, we had distinguished child and adult phases of learning, where the animat would randomly move through the maze in the child phase and build a network, while it would pursue its motivations in the adult phase (Butz et al, 2008;Shirinov & Butz, 2009). However, a child phase during which the animat is exploring its body in space with somewhat random movements is not realistic in a cognitive system.…”

Section: Continuous Learning and Behaviormentioning

confidence: 99%

Self-Organizing Sensorimotor Maps Plus Internal Motivations Yield Animal-Like Behavior

2010

Self Cite

View full text Add to dashboard Cite

This article investigates how a motivational module can drive an animat to learn a sensorimotor cognitive map and use it to generate flexible goal-directed behavior. Inspired by the rat's hippocampus and neighboring areas, the time growing neural gas (TGNG) algorithm is used, which iteratively builds such a map by means of temporal Hebbian learning. The algorithm is combined with a motivation module, which activates goals, priorities, and consequent activity gradients in the developing cognitive map for the self-motivated control of behavior. The resulting motivated TGNG thus combines a neural cognitive map learning process with top-down, self-motivated, anticipatory behavior control mechanisms. While the algorithms involved are kept rather simple, motivated TGNG displays several emergent behavioral patterns, self-sustainment, and reliable latent learning. We conclude that motivated TGNG constitutes a solid basis for future studies on self-motivated cognitive map learning, on the design of further enhanced systems with additional cognitive modules, and on the realization of highly adaptive, interactive, goal-directed, cognitive systems.

show abstract