Table 1 Online ErrP detection during a simulated and robot control (TPR: true positive rate, TNR: true negative rate, bACC: balanced accuracy [(TPR + TNR)/2]).

From: Intrinsic interactive reinforcement learning – Using error-related potentials for real world human-robot interaction

Simulated robot scenario.
Training: observation task, Test: interaction task
Subject TPR TNR bACC
Subject 1 (female) 1.00 0.98 0.99
Subject 2 (male) 0.86 0.96 0.91
Subject 3 (female) 0.92 0.83 0.88
Subject 4 (male) 0.89 0.79 0.84
Subject 5 (male) 1.00 0.73 0.86
Subject 6 (male) 1.00 0.98 0.99
Subject 7 (female) 1.00 0.77 0.89
Mean ± SEM 0.95 ± 0.02 0.86 ± 0.04 0.91 ± 0.02
% CI 0.95 ± 0.06 0.86 ± 0.10 0.91 ± 0.06
Real robot scenario
Subject TPR TNR bACC
Subject 1 (female) 1.00 0.96 0.98
Subject 2 (male) 0.50 0.96 0.73
Subject 3 (female) 1.00 0.89 0.95
Subject 4 (male) 0.57 0.89 0.73
Subject 5 (male) 1.00 0.89 0.95
Subject 6 (female) 1.00 0.96 0.98
Subject 7 (male) 1.00 0.95 0.98
Average ± SEM 0.87 ± 0.09 0.93 ± 0.01 0.90 ± 0.04
% CI 0.87 ± 0.21 0.93 ± 0.03 0.90 ± 0.11
  1. Mean, standard error of mean (SEM), and 95% confidence interval (CI = mean ± margin of errors are reported. Note that the positive class stands for a wrong mapping (Err label, ErrP).