Mobile Robot Navigation with Incremental Self-Organizing Developmental Network Learned Based on Modified RRT
Mengchao Shi, Dongshu Wang · 2024
When a mobile robot performs tasks, it may encounter changeable unknown environments. When navigating in the changing unknown environment, mobile robot should be able to learn incrementally to gradually improve their navigation capabilities. Considering that the fixed network structure will limit the continuous learning ability of the robot, this paper proposes an incremental self-organized developmental network, which does not need to define the number of neurons in advance, and its memory capacity adaptively increases with the learned knowledge. The network is employed to learn continuously in multiple complex environments without forgetting previously learned knowledge and without neuronal redundancy or insufficiency. In addition, this paper integrates the Rapidly Exploring Random Tree (RRT) algorithm into autonomous continuous learning process of incremental self-organizing developmental network, using the modified RRT algorithm to perform a global search of the environment to obtain a feasible path that the network can learn. This method can target the information in the specific environment, reduce the sample data for training the developmental network, improve the learning efficiency, and the network can show better generalization ability after continuous learning. The experimental results validate the effectiveness and potential of the proposed network as well as its learning method.