• Robots

Research - Basal Ganglia

Action Selection by the Basal Ganglia

A selection problem arises whenever an animal has two or more behavioural alternatives which require the same mechanism for execution. The problem is: how does the animal chose between them? We recently proposed that the basal ganglia, a group of interconnected sub-cortical brain structures, play a crucial role in resolving the conflict. This was based on an analysis of the neurophysiological evidence of the basal ganglia's connections and configuration and their known roles in behavioural performance. Following the adaptive behaviour literature, we suggested that the basal ganglia are ideally placed to act as an action selection mechanism; in short, a central switching device for the vertebrate brain which routes motor programs to the appropriate motor plants and suppresses inappropriate actions. We also proposed that the basal ganglia would play a similar role for cognitive processes.

To support our proposal we have three ongoing research projects:

1. Computational models of the BG

Initially we developed a computational model of the intrinsic circuitry of the basal ganglia which showed that the basal ganglia are capable of action selection and switching between actions. Furthermore, the signal traces from this model match well with those of single cell recordings taken from individual basal ganglia nuclei. This model was extended to include anatomically inspired thalamocortical loops which enhanced its action selection performance. The extended model also allowed us to postulate roles for the thalamic reticular nucleus in action selection.

Currently, the modelling work is being developed in two directions. First, spiking model neuron versions of both the intrinsic and extended models are being developed to investigate the affects of spike timing on action selection and the possible sources of oscillation within the basal ganglia, which is implicated in Parkinsonian tremor. Secondly, the modelling work is being carried out at a lower level, including a detailed model of a micro-circuit of the basal ganglia, the subthalamic nucleus-globus pallidus pars externa loop, and a model of the interactions of projection neurons in the striatum (the primary input nucleus of the basal ganglia).

1. The computational model of the basal ganglia

2. Thalamic and cortical interactions with the basal ganglia

3. 'Direct pathway' connections to the globus pallidus

4. Biophysical models of the striatum and basal ganglia

2. Robots

The intrinsic and extended computational models of the basal ganglia have been embedded in a robot control architecture. The robot itself interacts with a real world environment and shows a variety of behaviours, including food-seeking and wall-following, which are analogous to rat behaviours. These behaviours are successfully selected and switched between by the embedded basal ganglia model and coherent sequences of behaviour are formed. This is the strongest evidence yet that the basal ganglia are directly involved in action selection.

1. Prescott, T. J., Montes-Gonzalez, F., Gurney, K., Humphries, M. D. and Redgrave, P. (2006). A robot model of the basal ganglia: Behavior and intrinsic processing. Neural Networks, 19, 31-61.

2. Dopamine modulation in a robot model of action selection by the basal ganglia (poster presented to the Society for Neuroscience, November 2003)

3. Cognition and the basal ganglia

To investigate the possible role played by the basal ganglia in behaviours which are not motor-based, e.g. decision making, the intrinsic computational model has been adapted and added to a neural network model of processing in the Stroop task, based on Cohen et al's (1990) model. The classic Stroop task requires participants to either name the colour or say the word when presented with a word in a specific colour. In the model, the basal ganglia acts to decide between saying the colour or the word based on the input from the pre-processing done in the Cohen et al. network.

The results from this model show that it can successfully predict the changes in reaction times to different forms of word-colour pairs and to each of the tasks. Furthermore, it improves on the Cohen et al.'s original network by correctly replicating the affect on response times of increasing stimulus onset asynchrony.