Better Living Through Chemistry: Evolving GasNets for Robot Control

By P. Husbands, Tom Smith, N. Jakobi, Michael O'Shea. 1998

This paper introduces a new type of artificial neural network (GasNets) and shows that it is possible to use evolutionary computing techniques to find robot controllers based on them. The controllers are built from networks inspired by the modulatory effects of freely diffusing gases, especially nitric oxide, in real neuronal networks. Evolutionary robotics techniques were used to develop control networks and visual morphologies to enable a robot to achieve a target discrimination task under very noisy lighting conditions. A series of evolutionary runs with and without the gas modulation active demonstrated that networks incorporating modulation by diffusing gases evolved to produce successful controllers considerably faster than networks without this mechanism. GasNets also consistently achieved evolutionary success in far fewer evaluations than were needed when using more conventional connectionist style networks.

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Citations in the corpus

Born to Learn: the Inspiration, Progress, and Future of Evolved Plastic Artificial Neural Networks (2017, methodology): “Popular algorithms applied to EPANNs are reviewed later in Section III-A.”
Diffusion-based neuromodulation can eliminate catastrophic forgetting in simple neural networks (2017, methodology): “GasNets evolve the location, connectivity, and diffusion parameters for diffusing nodes [71].”
On the design of neural networks in the brain by genetic evolution. (2000, methodology): “…et al., 1996; Huber et al., 1996; Lund and Parisi, 1996) to behavioural robotics (Floreano and Mondada, 1994; Hoshino and Tsuchida, 1996; Kuwana et al., 1996; Husbands et al., 1998), with the genetic algorithm used as a tool to discover neural networks capable of solving certain kinds of task.”
Neuroevolution: from architectures to learning (2008, result): “This is in line with Magg and Philippides [43] whose results indicate that GasNets perform extremely well on tasks that require some kind of neural pattern generator or timer, while performing worse on tasks which do not require different time scales in the network.”
Evolving spike-timing-dependent plasticity for single-trial learning in robots (2003, background): “Additional elements, such as the directional damping of synaptic change (Rubin et al. 2001) or longer-term stabilizing regulation based on post-synaptic activity ( Horn et al. 1998; Turrigiano 1999), may come to the rescue.”
The road to everywhere : Evolution, complexity and progress in natural and artificial systems (2008, unknown): “For example, in GasNets [86], each neuron specifies its connection by defining “branching zones”, which correspond to portions of a circular area in a 2D plane.”
Neuronal Plasticity and Temporal Adaptivity: GasNet Robot Control Networks (2002, background): “For further details see Husbands et al. (1998) and Smith (2002).”
Selectionist and Evolutionary Approaches to Brain Function: A Critical Appraisal (2012, methodology): “See Husbands et al. (1998, 2010) for further details.”
The evolvability of artificial neural net-works for robot control (2002, unknown): “(Husbands et al., 1998b).”
Neuromodulation of Reactive Sensorimotor Mappings as a Short-Term Memory Mechanism in Delayed Response Tasks (2002, background): “Concerning the relation to our work, similar comments apply as in the case of Husbands et al. (1998).”