Replicating Associative Learning of Rodents with a Neuromorphic Robot in an Open-Field Arena

Tianze Liu¹Kang Jun Bai²Hongyu An^1*

• ¹Michigan Technological University, Houghton, United States
• ²Air Force Research Lab, Rome, United States

This research emulates associative learning in rodents using a neuromorphic robot within openfield arena environments. The neuromorphic robot is constructed by deploying computational models of neural systems in a mobile robot for perception and navigation. Various coding schemes, including rate and population coding, are utilized for different perception signals. The simulations and experiments demonstrate that our neuromorphic robot successfully replicates the classic associative learning experiments of rodents by memorizing the causal relationship between visual cues and other favorable or unfavorable stimulus locations in an open-field arena. The learning process in our neuromorphic robot relies on Hebbian principles and synaptic plasticity. This work has several unique contributions. Firstly, the neuromorphic robot offers a new embodied simulation platform for studying memory and learning research that does not rely on animal models. By embodying computational models within a physical robotic platform, our neuromorphic robot enables the study of spatial memory and cognitive processes in a real-world context. Secondly, the self-learning capability of our associative learning model enables the neuromorphic robot to explore and learn from their experiences rather than large-scale labelled datasets. In our associative learning model, which includes fewer neurons (19 neurons) while conducting a functional self-learning capability. Moreover, we studied the principles of how to determine the synaptic weights and threshold voltage of neurons, showing the design guidelines for more complicated associative learning model developments.