In this paper, learning characteristics in a robot manipulation task are discussed. Firstly, the learning phenomenon is analysed by means of the logarithmic linear regression model considering the confidence interval.
Based on this model, the relationship among characteristics in the learning process (learning rate, performance, confidence intervals in the learning curve at the first and the last trial, values of learning curve at the first trial and the last trial, dispersion ratio) are clarified. Moreover, the relevance between the learning effects and the performance is pointed out.
As a result, the strong positive correlation between the dispersion ratio and the learning rate, and the significant positive correlation between the initial confidence interval and the final confidence interval were found out. Moreover, there was no significant correlation between the learning effect and the performance.
View full abstract