Amygdala Modeling with Context and Motivation Using Spiking Neural Networks for Robotics Applications
Abstract
Cognitive capabilities for robotic applications are furthered by developing an artificial amygdala that mimics biology. The amygdala portion of the brain is commonly understood to control mood and behavior based upon sensory inputs, motivation, and context. This research builds upon prior work in creating artificial intelligence for robotics which focused on mood generated actions. However, recent amygdala research suggests a void in greater functionality. This work developed a computational model of an amygdala, integrated this model into a robot model, developed a comprehensive integration of the robot for simulation, and live embodiment. The developed amygdala, instantiated in the Nengo Brain Maker environment, leveraged spiking neural networks and the semantic pointer architecture to allow abstraction of neuron ensembles into high-level concept vocabularies. Test and validation were performed on a TurtleBot in both simulated (Gazebo) and live testing. Results were compared to a baseline model which has a simplistic, amygdala-like model. Metrics of nearest distance and nearest time were used for assessment. The amygdala model is shown to outperform the baseline in both simulation, with a 70.8% improvement in nearest distance and, 4%improvement in nearest time, and real applications with a 62.4% improvement in nearest distance. Notably, this performance occurred despite a five-fold increase in architecture size and complexity.
Document Details
- Document Type
- Technical Report
- Publication Date
- May 01, 2022
- Accession Number
- AD1169577
Entities
People
- Matthew A. Zelgen
Organizations
- Wright State University