Haptic exploration of objects usually consists of repeated exploratory movements and our perception of their properties is the result of the integration of information gained during each of these single movements. The serial nature of information integration in haptic perception necessities that perceptual estimates from single exploratory movements are retained in memory. Here we propose an optimal model for serial integration of information in haptic perception considering memory limitations by extending a previously proposed model to more natural explorations. We test the model by predicting discrimination performance in free and restricted explorations of softness. Our model predicts well the order of performance given different exploratory patterns in both, free and restricted explorations and the magnitude of performance in free explorations, suggesting that integration of information at least in natural haptic exploration can be well approximated by optimal integration given memory limitations. We further used the predictions of the model to assess whether participants adjust their exploration to maximize performance, by e.g. using an optimal switching strategy. Our results suggest that given the trade of between the improvement of performance and muscular costs of switching, participants invest additional muscular costs only as long there is maximal gain in performance.