We attempted to predict how participants would adapt to different dynamics which were experienced over the same physical states, based on normative predictions for how par-ticipants might chunk motor plans at baseline. Therefore, we predicted that pre-exposure chunking would scaffold chunking when dynamics changed. To test this, we compared central target dwell and response times across null and curl field dynamics, and the degree to which participants were able to concurrently adapt to opposing perturbations.
Across participants in all groups, dwell times at the end of the pre-exposure phase correlated with those at the start of exposure (Figure 5.5B, r = 0.55, p = 0.0011, slope of 0.67),
B
Pre-exposure Early exposure
Dwell time (s)
Late exposure
Early exposure dwell at central target (s)
A
Group 1 Group 2 Group 3 Group 4
Fig. 5.5 Central target dwell time in follow-through movements did not depend on trial set. A) Central target dwell time on unconstrained (null field) follow-through movements during the final 40 trials of the pre-exposure phase (lightest colour), on perturbed (curl field) follow-through movements during the first 40 trials of the exposure phase (middle colour), and the final 40 trials of the exposure phase (darkest colour) for each trial set. Model predictions for motor chunking in familiar dynamics were not reflected in these dwell times.
Bars show mean ± s.e. across participants. B) Central target dwell times at the end of the pre-exposure phase (final 40 trials) were correlated with dwell times at the start of the exposure phase (first 40 trials). Dots show individual participants.
suggesting a tendency for participants to adapt an existing chunking scheme, rather than adopt a completely new one when switching from baseline to perturbed environmental dynamics. However when dwell times across groups for different phases of the experiment were compared, a mixed-effects ANOVA with a fixed effect of epoch (three levels: 40 trials in each of pre-exposure, early exposure and late exposure), a fixed effect of group and a random effect of participants showed that there was a main within-subjects effect of epoch (Figure 5.5A, mixed-effects ANOVA, F2,56= 4.55, p = 0.015), although there was no between-subjects effect of group (mixed-effects ANOVA, F3,56= 1.98, p = 0.14) or an interaction (mixed-effects ANOVA, F6,56= 0.60, p = 0.73). Therefore while individual participants’ dwell times were correlated when transitioning from unperturbed to perturbed dynamics, this change in dynamics did affect how long participants waited at the central target before following through. Specifically, post-hoc reduced mixed-effect ANOVAs revealed that participants increased their dwell times when the dynamics initially changed (mixed-effects ANOVA, F1,28= 6.41, p = 0.017), and subsequently decreased dwell times as the exposure phase continued (Figure 5.5A, mixed-effects ANOVA, F1,28= 6.67, p = 0.015).
For all groups, the time taken to initiate a response decreased across the experiment (Figure 5.6A and B; mixed-effects ANOVA, main effect of epoch, F2,56= 22.8, p = 5.7e − 8; no
Pre-exposure Early exposure Late exposure
Δ initiation time (s) (full follow through - centre only trials) D
Group 1 Group 2 Group 3 Group 4 cue to move
6 Block 156 6 Block 156 6 Block 156
B
Fig. 5.6 Movement initiation times for centre only and follow-through trials in each trial set. A) Initiation times in follow-through trials, which could be planned as either cohesive or segmented movements, and B) initiation times to central target only trials, which can only be planned as a single chunk. Group 1 experienced only follow-through trials.
Data shows mean ± s.e. across participants for pairs of blocks and includes just the final 4 blocks of the pre-exposure phase. C) Initiation times for all movements performed by individual participants across three different epochs, separated for each group. Epochs shown are for null field follow-through movements during the final 40 trials of the pre-exposure phase (lightest colour), for perturbed (curl field) follow-through movements during the first 40 trials of the exposure phase (middle colour), and for the final 40 trials of the exposure phase (darkest colour). Lines show individual participants; bars show mean ± s.e. across participants. D) The difference in response times between follow-through and central target only trials show that participants consistently waited longer before initiating a movement just to the central target. Movement response times became marginally more similar between trial types as the exposure phase continued. Data show mean ± s.e. across participants for pairs of blocks and includes just the final 4 blocks of the pre-exposure phase. Grey shaded blocks show the exposure phase.
effect of group, F3,56= 0.41, p = 0.74; no interaction, F6,56= 0.92, p = 0.49). This decrease in response time could reflect improvements in timing the response with the predicted go cue, or the learning of a direct stimulus-response association between the visual target stimuli and a pre-computed policy (Haith and Krakauer, 2018; Hardwick et al., 2017b), rather than reflecting the computation of a less complex motor plan. To isolate changes in motor planning time, from generic reductions in reaction time with practice, we also compared how response times changed across the exposure phase for central target trials (which can only be planned in a single element), versus follow-through trials (which can be planned elementally or cohesively), (Figure 5.6C). Unexpectedly, despite requiring a greater, or least equal planning depth, trials requiring follow-through movements took less time for participants to initiate a response to, than central target-only trials (Figure 5.6A and D). The difference in response times between the two trial types decreased across the experiment, such that, while both movements came to be initiated more rapidly, they also came to be initiated at more similar times to each other (Figure 5.6C, mixed-effects ANOVA, main effect of epoch (F2,42= 6.0, p= 0.0051), no effect of group (F2,42= 1.41, p = 0.27) and an interaction (F4,42= 3.08, p= 0.026).