TY - GEN
T1 - A biomimetic robot controller based on minimizing the unpredictability of the environment
T2 - 11th European Conference on Artificial Life, ECAL 2011
AU - Sánchez-Fibla, Martí
AU - Duff, Armin
AU - Bernardet, Ulysses
AU - Verschure, Paul F.M.J.
PY - 2011
Y1 - 2011
N2 - Rodents are optimal real-world foragers that regulate internal states, such as security, arousal, energy, etc., maintaining a dynamic stability with their surroundings. Free exploration is an interesting scenario as rodents display behavioral patterns that are very different from being random, even in the absence of reward. Our aim is to understand foraging behavior by implementing an artificial rat that behaves as real ones do. We depart from the hypothesis that rodents, when performing free exploration, may be minimizing the unpredictability of the environment in terms of internally mapping its structure and discovering all the actions that it can afford. This drive for exploration is counterbalanced by a drive for security. Building from a self-regulation model based on the Distributed Adaptive Control architecture (DAC), we implement a biomimetic control that uses this predictability principle to generate behavior. We validate the controller by solving a benchmark task in which the agent learns to displace a movable obstacle to discover unexplored areas of an arena.
AB - Rodents are optimal real-world foragers that regulate internal states, such as security, arousal, energy, etc., maintaining a dynamic stability with their surroundings. Free exploration is an interesting scenario as rodents display behavioral patterns that are very different from being random, even in the absence of reward. Our aim is to understand foraging behavior by implementing an artificial rat that behaves as real ones do. We depart from the hypothesis that rodents, when performing free exploration, may be minimizing the unpredictability of the environment in terms of internally mapping its structure and discovering all the actions that it can afford. This drive for exploration is counterbalanced by a drive for security. Building from a self-regulation model based on the Distributed Adaptive Control architecture (DAC), we implement a biomimetic control that uses this predictability principle to generate behavior. We validate the controller by solving a benchmark task in which the agent learns to displace a movable obstacle to discover unexplored areas of an arena.
UR - https://direct.mit.edu/isal/proceedings/ecal2011/23/106/110794
UR - http://www.scopus.com/inward/record.url?scp=84455161221&partnerID=8YFLogxK
U2 - 10.7551/978-0-262-29714-1-ch106
DO - 10.7551/978-0-262-29714-1-ch106
M3 - Conference publication
AN - SCOPUS:84455161221
T3 - Artificial Life Conference Proceedings
BT - ECAL 2011
PB - MIT Press Journals
Y2 - 8 August 2011 through 12 August 2011
ER -