We present a closed-loop neurorobotic system to investigate haptic discrimination of Braille characters in a reading task. We first encode tactile stimuli into spiking activity of peripheral primary afferents, mimicking human mechanoreceptors. We then simulate a network of second-order neurones receiving the primary signals prior to their transmission to a probabilistic classifier. The latter estimates the likelihood distribution of all characters and uses it to both determine which letter is being read and modulate the reading velocity. We show that an early discrimination of the entire Braille alphabet is possible at both first and second stages of the somatosensory ascending pathway. Furthermore, 89% of the characters are correctly recognised in a constant-velocity reading task, while a closed-loop modulation of the speed allows for faster scanning and movement kinematics similar to the ones observed in humans –though with a lower classification rate.