Fig. 8.

The ATTC method. The ATTC method has been developed based on the Therrien16 model. The ATTC method uses four trial sequences to estimate variability from. The learner only explores post-failure (i.e., ${\eta }^{\left(t\right)}\leftarrow N\left(0,{\sigma }_{\eta \ast }^2\right)$; otherwise ${\eta }^{\left(t\right)}=0$). Post-success, the learner learns by updating the aim point (horizontal lines) with a fraction α of the rewarded exploration. In the figure, motor noise is not displayed (${\sigma }_m^2$ = 0). In the ATTC method, the reference trial (in bold) has been changed from the (non-)successful trial to the trial preceding it. Exploration is estimated twice, by subtracting variability estimates calculated from trial-to-trial changes of trial sequences post-single failure and post-double success pairwise, depending on reference trial exploration (a, b vs. c, d). A weighted average of the two is used to obtain one exploration estimate. a, c Trial sequences ending with a single failure (t – 1). If trial t-3 was non-successful (a), trial-to-trial change is a difference between two exploration draws: exploration on the target trial and a fraction (1 – α) of the reference trial exploration. If trial t – 3 was successful (c), trial-to-trial change only consists of exploration on the target trial. b, d. Trial sequences ending with double success to calculate post-success variability from. If the trial t – 3 was non-successful (b), trial-to-trial change is a fraction (1 – α) of the reference trial exploration. If the first trial was successful (d), trial-to-trial change is zero