Constructive incremental learning from only local information
Neural Computation
Evolving neural networks through augmenting topologies
Evolutionary Computation
| Hi-index | 0.00 |
Neural networks are a popular technique for learning the adaptive control of non-linear plants. When applied to the complex control of android robots, however, they suffer from serious limitations such as the moving target problem, i.e. the interference between old and newly learned knowledge. However, in order to achieve lifelong learning, it is important that robots are able to acquire new motor skills without forgetting previously learned ones. To overcome these problems, we propose a new framework for motor learning, which is based on consolidation. The framework contains a new rehearsal algorithm for retaining previously acquired knowledge and a growing neural network. In experiments, the framework was successfully applied to an artifical benchmark problem and a real-world android robot.