Repetita iuvant: repetition facilitates online planning of sequential movements

Beyond being essential for long-term motor-skill development, movement repetition has immediate benefits on performance, increasing the speed and accuracy of a second execution. While such repetition effects have been reported for single reaching movements, whether they apply to movement sequences, or whether they affect planning or execution processes, has yet to be determined. We addressed these questions in two behavioral experiments using a discrete sequence production (DSP) task in which human volunteers had to perform short sequences of finger movements. In Experiment 1, participants were presented with random sequences and we manipulated whether participants had to execute the sequence (Go), or not (No-Go), and whether the sequence was repeated on the next trial. We establish that repeating a sequence of movements led to immediate improvements in speed without associated accuracy costs. The biggest benefit was observed in the middle part of a sequence, suggesting that the repetition effect likely results from facilitated online planning. This claim was further supported by Experiment 2, in which we kept a set of sequences fixed allowing participants to develop sequence-specific learning: once learning reduced the need for online planning, the benefit of repetition disappeared. Finally, we found that repetition-related improvements only occurred for the trials that had been preceded by sequence production, suggesting that action selection and sequence pre-planning may not be sufficient to reap the benefits of repetition. Together, our results highlight the importance of motor practice for enhancing our ability to link individual sequence elements into skilled sequential behavior. Significance Statement Even for overlearned motor skills such as reaching, repeating the same movement improves subsequent performance. How exactly brain processes associated with motor planning and execution might benefit from repetition, however, remains unclear. Here we report the novel finding of repetition effects for sequential movements. Our results indicate that this benefit of repetition is tied to faster and more accurate online planning of upcoming sequence elements. We also highlight how recent movement experience appears to be required to observe the repetition effect, suggesting that actual practice might be more beneficial to the human sensorimotor system than mental rehearsal for producing short-term performance improvements.


Introduction
. The discrete sequence production (DSP) task. A. Exp. 1 example trial: a sequence cue (white numbers on the computer screen) is followed by an execution cue (outline changes color, numbers are masked). Online visual feedback about key-presses was given during the response window (green asterisks for correct presses, red for incorrect presses), followed by reward points depending on performance. 30% of the trials in a block were No-Go trials (red outline + low-pitch sound, top), 70% were Go trials (green outline + high-pitch sound, bottom). B. The next trial could be either a Repetition of the same sequence (0.5 probability), or a Switch to a new sequence. C. Example trial in Exp. 1 with the following trial timing: preparation phase: 2.5 sec; response window: 2 sec; ITI: 0.5 sec. Dashed horizontal line indicates force threshold (1 N) to determine the moment of each key-press and release (dotted lines). P1 = press of first key; R4 = release of fourth key; IPI 1 = first inter-press interval. Total time (TT) = RT + MT. D. Exp. 2 design: repeating sequences, trial structure and timing. Note that the go-signal here is given via white box around the sequence cue. Task. We used a discrete sequence production (DSP) task in which participants were required 150 to produce sequences of key-presses with the five fingers of their right hand (Fig. 1A). Each 151 sequence was cued by 4 numbers ranging from 1 to 5, instructing which fingers had to be 152 pressed (e.g., 1 = thumb, 2 = index, … 5 = little). The sequence had to be executed by 153 sequentially pressing the fingers corresponding to the numbers on the screen, from left to right. 154 On each trial, participants were presented with a 4-item sequence and asked to prepare for the 155 corresponding finger presses (preparation phase). After a fixed delay of 2.5 seconds, an audio-156 visual execution cue would mark the beginning of the sequence response window (a fixed 2 157 seconds). On Go trials, the execution cue was a green frame accompanied by a high-pitch tone 158 (Fig. 1A, bottom), indicating that participants had to perform the planned sequence of finger 159 presses as quickly and accurately as possible (Go condition). On other trials, the execution cue 160 was a red frame accompanied by a low-pitch tone (Fig. 1A, top), instructing the participants to 161 remain as still as possible without pressing any key until the end of the response window (No-162 Go condition). To encourage sequence pre-planning before the Execution cue, at the beginning 163 of the response window the sequence cue was replaced by 4 asterisks masking the numbers. 164 Moreover, the sequence of key-presses had to be completed within 2 seconds from the 165 execution cue (timeout error after that). With each key-press, the corresponding asterisk turned 166 either green (correct press) or red (wrong press). Performance was evaluated in terms of both 167 execution speed and press accuracy. Speed was defined in terms of total time (TT), which 168 consisted of the reaction time (RT: from the onset of the Sequence cue to the first key-press) plus the movement time (MT: from the onset of the first key-press, P1, to the release of the last 170 key-press, R4). A single press error invalidated the whole trial, so accuracy was calculated as 171 percent error rate (ER) per block of trials (number of error trials / number of total trials x 100). At 172 the end of the response window, during the 500 ms inter-trial interval (ITI), participants were 173 presented with performance points appearing in place of the asterisks.

175
Feedback. To motivate participants to improve in speed (TT = RT + MT) and accuracy (1 -ER) 176 of sequence production, we gave participants performance feedback on each trial. The 177 performance score was based on the following point system: -1 points for timing errors (i.e., 178 anticipation of the execution cue, or movement initiation in No-Go trials); 0 points for correct 179 timing but wrong finger press (any one wrong key-press); +1 points for correct timing and press 180 (i.e., movement initiation in Go trials, or no movement in No-Go trials); and +3 points for correct 181 timing, correct press, and TT 2% or more faster than TT threshold. TT threshold would decrease 182 by 2% from one block to the next if both of the following performance criteria were met: median 183 TT in the current block faster than best median TT recorded hitherto, and mean ER in the last 184 block < 25%. If either one of these criteria was not met, the thresholds for the next block 185 remained unchanged. At the end of each block of trials, the median TT, mean ER, and points 186 earned were displayed to the participants. At the end of the session, monetary compensation 187 corresponded to the amount of performance points accumulated (points < 750 = 10 $; 750 ≤ 188 points < 1000 = 12 $; points ≥ 1000 = 15 $).

189
Penalizing timing errors (-1 points) more than press errors (0 points) might have made 190 participants more cautious and increased their RTs. Thus, to encourage full preparation of the 191 sequence, for the last 20 participants of Exp. 1 we gave equal weight to timing and press errors 192 (both 0 points). Successive analyses showed no difference in performance between the two 193 groups (with or without penalty for timing errors), suggesting that participants adopted a similar 194 strategy regardless of the penalty.
The scoring system in Exp. 2 was identical to the one in Exp. 1 without any additional 196 penalty for eventual timing errors (0 points). Furthermore participants in Exp. 2 were paid a flat 197 hourly rate (7 $), regardless of the specific amount of points accumulated.  Moreover, to keep sequences of a similar level of difficulty, we removed from the permutation 216 pool all sequences in which any number repeated (i.e., each number could only appear once 217 per sequence), or that included "runs" (more than 2 fingers in either increasing or decreasing 218 order; e.g., 1-2-3, or 3-2-1).

Exp. 2.
To explore how sequence learning affects the repetition effect, we designed a second 221 experiment where the set of sequences remained fixed over two consecutive days of testing 222 following overnight consolidation. We used 8 4-item sequences including all fingers of the right 223 hand except for the ring finger. The sequences were selected according to the following criteria: 224 1) each finger was used only once per sequence; 2) each finger started 2 of the 8 sequences; 3) 225 each finger was pressed in every ordinal position twice across sequences; and 4) no more than 226 2 neighboring fingers pressed in a row (i.e., as in Exp. 1, we excluded "runs").

227
In contrast to Exp. 1, Exp. 2 did not contain any No-Go trials and the preparation phase 228 was shortened to a fixed 1 s. Also, the execution cue was presented only visually (white box 229 around the sequence cue), the sequence cue was not masked, and the duration of the response 230 window was not fixed (i.e., TT dictated the actual duration of the trial, with the ITI occurring right 231 after the last key-press). Finally, sequence repetition was not randomized, but counter-balanced 232 across sequences. Each sequence was executed from a minimum of once (i.e., a Switch) to a from 4 % to 2.9 % (paired-samples t-test t44 = 2.777, p = 0.008).

297
Overall, our results suggest that repetition of a sequence improves both the initiation of a 298 pre-planned movement, as well as the speed by which the repeated sequence can be 299 performed. The accuracy advantage proved that this effect did not arise at the expense of 300 reduced execution accuracy.

302
Repetition benefit on MT arises from improved online planning, not execution processes 303 The results so far indicate that sequence repetition improves initiation (RT) and movement (MT).

304
Should this be taken as an indication that repetitions improve execution-related, rather than 305 planning-related processes? Not necessarily so. In a previous study we have demonstrated that 306 sequence MT (the time from first to last key-press) is not only a function of motoric processes, 307 but is also strongly influenced by the speed of online planning (Ariani and Diedrichsen, 2019).

308
Even for short sequences, only the first 2-3 key-presses can be fully pre-planned, whereas later 309 movements appear to be planned online, that is during of the execution of the beginning of the To examine this issue, we inspected the 3 inter-press intervals (IPIs) between the onsets 315 of the 4 key-presses separately. The second transition was the slowest, while the first and last 316 transition were nearly equally fast (Fig. 3A). This indicates a "2-and-2" rhythm, in which each 4-317 item sequence begins with two quick presses, followed by a brief pause, and then again by two 318 quick presses. Given that the sequences changed randomly from block to block, all possible 319 finger transitions could occur with equal probability at each position of the sequence. Therefore, 320 this effect cannot be explained by biomechanical factors (e.g., some transitions being harder than others). Instead, the pattern or results suggests a clear influence of online planning: the 322 first two key-presses can be fully pre-planned and can therefore be executed quickly; then 323 execution needs to slow down until online planning of the remaining two key-presses is finished. No-Go trial of the same sequence, we restricted this analysis to the first repetition of a sequence 348 (i.e., max two executions in a row). 349 We found that the repetition effect on RT was significant when the previous trial had and previous trial type (F1,44 = 4.898, p = 0.032). As excepted, the effect was also visible when we split up the MT into IPIs (Fig. 4C)

412
In Exp. 1, sequences varied randomly from block to block, effectively preventing participants 413 from learning a specific set of key-press transitions. Therefore, it is likely that overall 414 performance improvements largely reflected sequence-general learning (Fig. 6A) 481 In a previous study we found that benefits of increased preparation reach an asymptote after 482 ~1.5 seconds (Ariani and Diedrichsen, 2019). Thus, by using a delayed-response paradigm, we 483 can be relatively confident that our RT measure mainly reflects the initiation of a pre-planned 484  Improved sequence production is consistent with more efficient online planning 495 Repetition accelerated not only RT, but also MT for repeated sequences. Critically, a detailed 496 analysis of the inter-press-intervals revealed that the transition between the first two key-497 presses, which was likely fully pre-planned, was not influenced by the repetition. Rather, the   Is pre-planning sufficient to drive the repetition effect, or is movement required? 532 The repetition effect was only present when the sequence was actually executed on the 533 previous trial. Sequence pre-planning alone did not appear to be sufficient for the repetition 534 effect. This was slightly surprising given the notion that neural states for planning and execution

543
A potential explanation for our finding could be that, despite having enough time and 544 information, participants did not fully pre-plan the response during the preparatory period.

545
Indeed, reaction times were relatively long (~400 ms) for triggering a pre-planned sequence.

546
Perhaps participants used time after the go-signal to complete preparation. In this view, the 547 completion of pre-planning would be essential for the subsequent repetition benefit.

548
Nonetheless, our experiment was designed to motivate participants to pre-plan the sequence in 549 advance during the delay. We masked the sequence cue so that they could not rely on that after 550 the go-signal. Go and No-Go trials were pseudo-randomly ordered, and we included a higher 551 proportion of Go trials (70%), such that, more often than not, participants would have to act on 552 the pre-planned sequence. Finally, we rewarded participants on the sum of RT and MT, 553 meaning that an easy way to earn more money would be to shorten RTs. Thus, it is hard to see 554 how more complete pre-planning could be achieved.

555
Rather, a more likely explanation of our result is that the repetition benefit arises only 556 when the planning trajectory for the entire sequence is complete, something that requires online 557 planning (Fig. 7B). According to this view, neither the pre-planning of the initial part of the 558 sequence, nor the motoric execution of the individual sequence elements would be enough to 559 produce a behavioral advantage with repetition. Instead, it is revisiting the whole trajectory in the 560 preparatory state-space that improves subsequent RTs and MTs.