Robots are an essential part of our industrial production process and in the future robots might accompany us every day.

In all this we need the computing of movement trajectories to solve a task, including considering constraints. This task solving is condensed in the term planning and thus planning is a fundamental knowledge in nearly all robotic tasks. For this application Spiking Neural Networks (SNN) are one of the new state of the art learning algorithms. SNNs incorporate the concept of time, they also include information to be processed asynchrony, event-based and without delay.

SNN’s are uniquely equipped to solve planning problems. This demands an increase of synapses, which is (going to be) compensated by neuromorphic hardware. The number of synapses is significantly higher, this also reflects on the increased consumption. In this work, we want to show that models for synapse pruning decrease the synapses needed and in addition result in a significant decrease of memory consumption.