A vision-driven model of hippocampal place cells and temporally asymmetric LTP-induction for action learning

摘要

We describe a hippocampal neural model in which spatio-temporal features of the environment are extracted by visually driven neurons. The neuronal firing activity implicitly measures properties like agent-landmark distance and egocentric orientation to visual cues. This leads to a neural representation where populations of place cells encode spatial locations within the environment. In addition, temporally asymmetric long-term potentiation of synapses between place cells is used to learn a vector field representation providing a navigational map. We present experimental results obtained by testing our model with the mobile Khepera robot.