Fig. 3From: Automated calibration of somatosensory stimulation using reinforcement learningPerformance of the RL-based algorithm, BFA algorithm, expert user and naïve user. The results during the first day of characterization are shown. These plots are computed for: A final charge released by the stimulation parameters found, divided by low level and high-level calibration, comparing expert and RL performance. The RL was then compared to the expert for both low and high level in terms of percentage of improvements. B Time needed to perform the characterization of the nerve, C number of stimulations delivered and D overall sensation quality of the mapping. The bar plots represent the mean values and standard deviation of the measurements of 15 nerves of five independent subjects for the 4 approaches (p < 0.0083 (*), p < 0.0017 (**), p < 0.00017 (***)). The four conditions were then compared to the expert and expressed as a percentage of the expert performance. The scatter plots represent a direct comparison between the RL and the expert for each trialBack to article page