Add-if-Silent Rule-Based Growing Neural Gas with Amount of Movement for High-Density Topological Structure Generation of Dynamic Object

摘要

In order to realize a super-smart society (Society 5.0) where humans and robots coexist, there is a need for a perceptual system that can recognize the environment quickly and flexibly in an environment that changes from moment to moment. In an unknown environment, the characteristics of objects cannot be known in advance, and thus prior learning-based recognition methods such as deep reinforcement learning may not be able to cope with this situation. In this study, we construct a 3D topological map of the environment in real-time using Growing Neural Gas (GNG), which can learn 3D topological structures even for unlearned objects. However conventional GNG have the problem that they cannot generate nodes with high-density for distant objects and cannot identify whether an unknown object is static or dynamic. Therefore, by directly adding useful input data as a new node (reference vector) based on the object category labels of the winner nodes (nearest nodes) to the input vector (3D point cloud), it is possible to generate high-density topological structures even for distant objects. We proposed the Add-if-Silent rule-based GNG with Amount of Movement (AiS-GNG-AM), which can identify between static and dynamic objects based on the past amount of movement of a node. The effectiveness of the proposed method is verified through experiments using a 3D dynamics simulator.