Authors: Orit Herzberg, Catherine S. Tamis-LeMonda, Margaret Shilling, Karen E. Adolph
Categories: Article, Locomotion, walking, transitions, falling, variability, home environment
Source: Developmental psychology
Doi: 10.1037/dev0002131
Authors: Orit Herzberg, Catherine S. Tamis-LeMonda, Margaret Shilling, Karen E. Adolph
Learning to walk—like learning any motor skill—requires practice. But what are the critical components of infants’ natural practice regimen? Traditionally, researchers focus on practicing the target behavior (e.g., days since walk onset, number of steps/hr). However, part of what makes skills truly functional is the ability to enter the task space (here, transitions from non-upright to upright postures), produce and pause the target behavior (transitions between standing and walking), and use the behavior in varied contexts (e.g., walking from one surface to another and from one place to another). Thus, we expanded the concept of practice to include moment-to-moment, self-generated transitions that require rapid adaptations to ever-changing affordances. During two hours of spontaneous home activities, regardless of age, infants (13-, 18-, and 23-month-olds, N = 12 per age; half girls, half boys; 72% White, 28% Black, Asian, or mixed race; 83% non-Hispanic, 17% Hispanic; 94.4% college educated; living in urban city) generated immense numbers of transitions/hr (total M = 642, range = 153–1179). Infants accumulated such large numbers of transitions for three (1) Infants continually pushed their limits; (2) bouts of walking, standing, sitting, and crawling were frequent (e.g., M = 206.2 walking bouts/hr) but brief in duration (e.g., M = 2.1 s per bout of walking); and (3) infants repeatedly revisited surfaces and places, regardless of home size and layout. Transitions are a unique form of variability that promote skill acquisition via practice adapting ongoing behaviors to variations in posture, tasks, and features of the environment.
Infants acquire a large repertoire of motor skills—walking, talking, manipulating objects, and so on—and each skill undergoes dramatic improvements. Practice promotes improvements, of course, but what kind of practice? The obvious answer is that self-generated practice performing the target skill is critical (Ericsson, Krampe, & Tesch-Romer, 1993; T. D. Lee & Schmidt, 2025; Newell, 1991). But perhaps the answer is not so obvious. What exactly does it mean to ‘practice a target skill’?
Consider infant walking. Infants’ natural practice regimen entails immense numbers of brief bouts of self-generated, time-distributed, variable, error-filled practice (Adolph & Hoch, 2019; Adolph, Hoch, & Cole, 2018). Each hour, toddlers average 2400–4200 steps and travel the distance of 8 American football fields (Adolph et al., 2012; Hoch, O’Grady, & Adolph, 2019).
Albeit immense, infants’ practice regimen is not like taking uniform, continuous steps on a motorized treadmill. Indeed, treadmill training—a popular intervention for infants with Down’s syndrome, spina bifida, and cerebral palsy, and for adults with spinal cord injury, stroke, and Parkinson’s disease—can strengthen legs and spinal pathways, but it is not equivalent to functional walking in the real world (e.g., Angulo-Barroso, Wu, & Ulrich, 2008; D.K. Lee & Sansom, 2019; Reisman et al., 2009). Why not? In treadmill training, people are placed or suspended upright, a motor turns the treadmill on and off, and the treadmill has only one fixed, uniform path direction, surface, and context. But in everyday life, people have to get themselves into an upright posture, start and stop at will, and navigate varied surfaces and layouts.
Indeed, natural practice is variable. Infants walk in short and long bursts (1 to 20+ steps) of variable speeds separated by longer rest periods (Adolph et al., 2012; Cole, Robinson, & Adolph, 2016; Hoch et al., 2019; Hospodar, Hoch, Lee, Shrout, & Adolph, 2021; D. K. Lee, Cole, Golenia, & Adolph, 2018). They travel in winding, circuitous paths (M = 73% of bouts); they take steps in all directions and in place (M = 81% of bouts not exclusively forward) (D. K. Lee et al., 2018); and they visit most available areas in a laboratory playroom (Hoch et al., 2019; Thurman & Corbetta, 2017). Errors are frequent—infants fall 17 times/hr (Adolph et al., 2012)—but falling does not discourage infants from practice. After a fall, infants hoist themselves upright and are back at play within 2 s (D. Han & Adolph, 2021). Formal modeling with simulated robots suggests that infants’ natural practice regimen—replete with variability and errors—facilitates improvements in functional walking skill (Ossmy et al., 2024; Ossmy et al., 2018). However, a critical component of variable practice is missing from this characterization.
We propose that moment-to-moment transitions are fundamental to infants’ natural practice regimen. Transitions introduce a unique form of variability that requires infants to instantly adapt ongoing behaviors to variations in posture, tasks, and features of the environment. Natural activities are always in flux. The everyday environment is a continually shifting landscape. Thus, walking—like talking, manipulating objects, and every other behavior—requires continual, moment-to-moment adaptations to ever-changing affordances (E. J. Gibson, 1988; J. J. Gibson, 1979) and moving between states of equilibrium and disequilibrium (Thelen & Smith, 1998; Thelen & Ulrich, 1991).
A functional practice regimen requires babies to enter the task space of the target behavior, produce and pause the target behavior at will, and use the target behavior in varying environmental contexts. In contrast to other forms of variability that are ‘stand-alone’ behaviors (such as path length and step direction), a transition, by definition, entails a switch between two temporally adjacent behaviors or two spatially adjacent environmental contexts (e.g., Marsala & VanSant, 1998; Thurman & Corbetta, 2020).
Transitions into an upright posture—entering the task space—set the stage for walking, and thus are an important item on many assessment instruments and ‘milestone’ charts (Bayley & Aylward, 2019; CDC US Center for Disease Control, 2025; Folio & Fewell, 2000; Marsala & VanSant, 1998; VanSant, 1988). Getting the body upright without pulling up on furniture or help from a caregiver requires greater leg strength and balance control than merely standing (imagine adults doing ‘squats’ to build leg strength and balance). Thus, most infants master independent transitions to upright after they can already walk (McGraw, 1940; Piper & Darrah, 2021). Lowering the body to the floor, perhaps to play, provides opportunities to practice getting back up. Likewise, falls inadvertently instigate transitions to upright so that infants can resume walking (D. Han & Adolph, 2021).
Transitions between walking and standing—producing and pausing the target behavior—are important for skill acquisition because each initiation and termination of gait is a lesson in creating disequilibrium to walk and recapturing equilibrium to stand (Assaiante, Woollacott, & Amblard, 2000; Breniere, Bril, & Fontaine, 1989; Bril & Ledebt, 1998; Ledebt, Bril, & Breniere, 1998). Moreover, self-generated transitions between producing and pausing the target behavior allow infants to control their own activities and goals.
Finally, environmental contexts (e.g., ground surfaces, spatial layout) put different biomechanical constraints on walking and therefore provide different affordances for balance and locomotion. Transitions from one ground surface to another provide practice coping with different demands on balance and propulsion (e.g., hard, slippery, wood floor vs. deformable, high-traction carpet). Such surface transitions can be treacherous for infant walkers who often slip and trip (Adolph, Joh, & Eppler, 2010; D. Han & Adolph, 2021; D. Han et al., 2023; Joh & Adolph, 2006). Transitions between different places (e.g., bedroom to living room) ensure practice with navigation and steering to cope with different spatial layouts and arrangements of furniture (E. J. Gibson, 1988; J. J. Gibson, 1979).
Here, we took seriously the question of what it means to practice a motor skill. We focused on walking because the behavior is easily observable, it undergoes dramatic improvements in the first few months after walk onset, and practice depends on infants’ self-generated input. We tested 13-, 18-, and 23-month-olds to span the period of early walking when improvements are most rapid and dramatic to a later period when improvements are more gradual and subtle (Adolph, Vereijken, & Shrout, 2003; Bril & Breniere, 1992, 1993; Bril & Ledebt, 1998; Hospodar et al., 2021). Whereas most prior work focused on spontaneous walking for short durations (5–20 min) in an unfamiliar laboratory playroom (e.g., Adolph et al., 2012; Hoch et al., 2019; D. K. Lee et al., 2018; Thurman & Corbetta, 2017), we observed infants for two hours (one hour on each of two days) during natural activity in their familiar home environment—the setting where presumably most everyday practice occurs.
The real novelty of our work is our focus on moment-to-moment transitions between behaviors and environmental contexts. Part of what makes walking truly functional is the ability to get upright from non-upright postures, start and stop walking at will, and navigate a varied environment. Thus, understanding the practice regimen for walking is incomplete without characterizing the critical transitions.
As shown in Figure 1 and in exemplar videos at databrary.org/volume/1109/folder/76131, we quantified four key types of self-generated (1) from non-upright to upright postures; (2) between walking and standing; (3) in walking from one surface to another, and (4) while moving between different places in the home. Because the number of transitions depends on the number of and criteria for different categories, we lumped behaviors into a small number of categories and used conservative criteria to prevent inflating the number of transitions. We collected data on home layout and room dimensions because differences in the environment affect possibilities for transitions among ground surfaces and spaces.
We also quantified how much infants walk and what infants did between bouts of upright activity—whether infants were constrained (when practice walking was impossible) or crawling or sitting (when they could decide to get upright and walk). Although group averages are important to characterize infant experience, we also reported individual data because each baby learns to walk—just as they learn any skill—based on their own practice regimen, not from group averages or another infant’s input (Vong, Wang, Orhan, & Lake, 2024). Finally, we tested whether individual differences in practice were stable across visits and whether practice varies by infant age, walking experience, and home size.
Videos were culled from a larger dataset collected between 2017 and 2020 and shared (with parents’ permission) with authorized investigators in the Databrary video library (databrary.org/volume/563). We analyzed the first hour of each two-hour home visit of walking infants. The study and analyses were not pre-registered. However, for transparency and reproducibility, illustrative videos, the video coding manual, Datavyu scripts, processed data, and analysis scripts are publicly shared at databrary.org/volume/1109. Datavyu coding spreadsheets are shared in the same volume with authorized Databrary investigators.
Families were recruited from the New York City area through hospitals, referrals, and brochures. From the shared dataset, we reused data from a subset of 36 infants, 6 boys and 6 girls in each of three age 13-month-olds (12.7 to 13.3 months), 18-month-olds (17.8 to 18.2 months), and 23-month-olds (22.8 to 23.2 months); Figure 2A. Based on “the first time mothers saw their infant walk across the room for 3 m without stopping or falling” (Adolph et al., 2003), time since walk onset ranged from 3 days to 12.3 months; Figure 2A. Older infants had more months walking, r(34) = .937, p < .001.
All infants were firstborn singletons, and all were healthy and born at term. Parents reported infants’ race as White (72%), Black (3%), Asian (6%) or mixed race (19%) and their ethnicity as non-Hispanic (83%) or Hispanic (17%). In most (65%) visits children wore lightweight clothing (pajamas, one layer of light clothes, t-shirt and leggings, or onesie). Families received $150 for participation.
Most families lived in apartments (86.6%) rather than single-family homes. Only 6 homes were multi-level with stairs. Based on video home tours (databrary.org/volume/563) and laser measurements of room dimensions, we used architectural design software (Home Designer Pro 2021) to digitally model the layout of the home—including furniture and large objects on the floor—to calculate home size and open floorspace where infants could move.
Figure 3 shows the smallest (44.2 m^2^) and largest homes (178.8 m^2^), M = 77.8 m^2^. Furniture occupied M = 34.4% of the floor space overall (range = 33.9% to 53.3%), but every room (except hallways) had furniture that required steering. Home size was correlated with open area, r(34) = .986, p < .001, so we analyzed open space because it was more conceptually meaningful. Most homes (60.6%) had 2 bedrooms, 34.3% had 1 bedroom, and 6.1% had 3; most homes (72.2%) had 1 bathroom, 25% had 2 bathrooms, and 2.8% had 3+ bathrooms; most (63.9%) had a separate kitchen area (surrounded by 3+ walls), and the others had a combined living/kitchen area.
A researcher video recorded infants in two home visits, between infant mealtimes and naps. Second visits were scheduled M = 5.3 days apart (range = 1 to 10 days). Because family activity fluctuates between weekdays and weekends and across the day, visits were scheduled on two weekdays at approximately the same time of day (M difference in time = 1.1 hr; range = 0 to 6.5 hr). Mothers were instructed to “go about their natural activities and to ignore the researcher.” They were free to play with their infants, watch TV, do household chores, work on their computer, and so on, and most mothers did a range of activities. The researcher recorded infants’ behaviors with a handheld video camera (30 fps) and did not interact with infants or mothers.
We annotated infant behaviors using Datavyu (datavyu.org) software that time locks user-defined events to their frames in the video. We annotated behaviors in passes for efficiency (coding every behavior and context simultaneously is more laborious and cognitively taxing). We used conservative criteria to identify behavioral events (e.g., when walking ends and standing begins; when baby is sitting versus crawling) to avoid inflating the number of transitions. The data from multiple passes were integrated using scripts to parse the data into six mutually exclusive behavior categories (constraint, walking, standing, crawling, sitting, and falling), the surfaces and places where walking and standing occurred, and periods of missing data (Figure 1A).
An initial annotation pass identified periods of constraint, locomotion, falls, and missing data (Figure 1A top row). Time constrained was when infants were in a device (e.g., highchair), on an elevated surface (e.g., changing table), held in mother’s arms or lap, or mother briefly supported infant’s body weight (e.g., mother pulled standing child onto her lap). Onset was the first video frame when infant’s body touched the device or mother’s lap or mother lifted the baby. Offset was the first video frame when infant’s body returned to the ground.
Time in locomotion was when infants moved their whole body in any direction or stepped in place. Onset was the first video frame when infants began a step and offset was the first frame when infants stopped moving for at least 1 s. Note, this is a conservative criterion because periods of double support during walking rarely exceed 0.5 s (Cole et al., 2016). Falls occurred when infants lost balance, and their body impacted the ground. Missing data were video frames when infants were off camera or infants’ legs were occluded and coders could not annotate posture or locomotion.
In a second pass, coders marked infants’ posture as upright or non-upright (Figure 1A top row). Upright postures entailed keeping balance with two feet on the floor and hands off the floor, and non-upright postures were everything else (baby’s bottom or hands touching floor, leaning on furniture, etc.). Thus, ‘upright’ denoted independent control of hands-free balance, including squatting (see Figure 4A). Upright began when infants bore their weight solely on their feet and ended at the onset of a non-upright posture or mother picking them up.
We used Datavyu scripts to parse the two coding passes into four mutually exclusive categories (Figure 1A top row): upright locomotion (walking), upright stationary (standing), non-upright locomotion (e.g., crawling), non-upright stationary (e.g., sitting). For walking, we considered the utility of counting the number of steps/bout. Time in motion and number of steps were highly correlated (r =.95*, p* <.001) based on the first 10 walking bouts of each infant’s two visits, so we relied solely on bout duration (Supplemental Figure 1). For non-upright locomotion, infants were crawling on the floor or climbing or leaning onto furniture in 86% of non-upright locomotor bouts, and they bum-shuffled, cruised, or knee walked on the remaining 14% based on each infant’s first 10 non-upright locomotor bouts, so we termed non-upright locomotion as ‘crawling.’
However, for non-upright stationary bouts, infants could be sitting (with bum on floor, furniture, or heels), kneeling, quadruped (resting weight on hands while side-sitting with bum off floor, resting on hands/knees, or ‘down-dog’ posture), or lying on the floor or furniture (prone or supine)—behaviors which were relevant for our analyses of transitions. Infants could also be on two feet but leaning on furniture or mother (with hands, belly, or back) or supported by mother’s hands (e.g., mother supported infant’s balance without lifting infant from floor)—behaviors which were not relevant for our analyses because they did not contribute to transitions to upright. Thus, we further categorized each non-upright stationary bout into sitting, quadruped, lying, leaning, or supported based on the first 10 non-upright stationary bouts, and excluded leaning and supported non-upright bouts from further analyses. Of the relevant, non-upright stationary bouts, infants were sitting in 70%, quadruped in 25%, or lying in 5%, so we termed those as ‘sitting.’
Transitions to upright were unsupported changes (no pulling up on furniture or help from mother) between temporally adjacent non-upright and upright postures (e.g., sitting to standing, crawling to walking); see green hash marks in Figure 1A top row and Figure 4A. Likewise, transitions between walking and standing were based on temporally adjacent behaviors (red hash marks in Figure 1A top row and Figure 4B). We only analyzed independent transitions (e.g., if infant was sitting, and mother placed baby upright, it did not count as a transition, but if infant was sitting and stood up, it counted as a transition).
In a third pass, coders marked the surfaces infants touched with each foot while walking or standing (time sitting or crawling was excluded, Figure 1A second row, Figure 4C). Onset was the first video frame when a foot touched a new surface; offset was the last video frame of both feet on the surface. Thus, a surface bout could contain multiple bouts of standing and walking (e.g., baby took 5 consecutive walking bouts on wood floor), and conversely, one bout of walking could cross multiple surfaces (baby walked continuously from wood floor to carpet to linoleum).
Coders initially identified 14 floor surfaces (wood, laminate, concrete, tile, high/medium/low-pile carpet, foam mats, etc.). Infants also stepped/stood on non-floor surfaces (surfaces adults would not walk on such as cushions, beds, couches, chairs—and their mother). Unexpectedly, babies stepped/stood on objects (things that moved or created bumps underfoot such as toys, laundry, and books).
To avoid over-inflating the number of transitions, we conservatively grouped floor surfaces into hard (rigid, non-deformable, smooth, flooring such as wood, tile, and linoleum) and soft (non-rigid, deformable, porous, floor coverings such as carpet, rugs, and foam mats), and retained non-floor surfaces and objects as two additional types of surfaces. Because infants often walked or stood on multiple surfaces simultaneously (i.e., one foot on two surfaces, each foot on a different surface), these were categorized as mixed (Figure 4C).
We calculated surface transitions based on walking between different spatially adjacent surfaces (blue hash marks in Figure 1A second row). We counted only instances when infants generated the transition. Thus, if mother picked infant up from the rug and placed baby onto the wood floor, it did not count as a transition.
We considered the possibility that the interface between infants’ footwear and the floor surface would affect walking (e.g., socks vs. barefoot on wooden floor). However, infants were mostly barefoot (50% of visits) or wearing socks (27.8%) the entire visit. Sometimes infants wore shoes (8.3%) but they rarely changed footwear (13.9% of visits), so we did not include footwear in further analyses.
In a fourth pass, we categorized places based on the video home tour. To avoid over-inflating the number of place transitions, we counted each space enclosed by at least three walls as a different ‘place’ (Figure 1A third row, Figure 4D). Bedrooms included parent, infant, shared, and guest bedrooms; bathrooms included parent, infant, shared, and guest bathrooms; hallways included halls and foyers; living spaces included living room, family room, and combined living room-kitchen (open floor-plan homes); kitchens were only counted if separated from the living area with floor-to-ceiling walls; other rooms included office and laundry room (Figure 3 shows exemplar homes). Based on these criteria, the number of possible places to visit ranged from 4 to 16, M = 7.1.
We counted transitions from one place to another only if infants walked between places (time sitting or crawling in or from one place to another was excluded, gold hash marks in Figure 1A third row, Figure 3, Figure 4D). Thus, walking from the living space to the hallway and then back to the living space counted as two transitions. We did not count mothers carrying infants to a new place as a transition.
For each annotation pass, a ‘bulk’ coder annotated 100% of each video and a ‘spot-checker’ annotated a random 25% of each 20-min video segment from each visit, blind to the bulk coder’s annotations, to ensure inter-observer reliability. This method ensured that agreement was based on representative data from each infant (Herzberg, Fletcher, Schatz, Adolph, & Tamis-LeMonda, 2022).
Inter-observer reliability was high for each pass. Overall percent frame agreement for time constrained, walking, standing, crawling, and sitting was 94.1%, κ = .92. Correlations between coders for the accumulated number of falls and bouts of walking, standing, crawling, and sitting were rs(34) ≥ .89, ps < .001. For annotations of surfaces, percent frame agreement was 95.6%, κ = .93. For annotations of places, a second coder ensured that the labeling of the places infants visited matched the house maps by watching the home visits and home video tours.
Coders reviewed disagreements after every few sessions. Although the number of disagreements was small, typos and careless errors were corrected to avoid propagating known errors into the final analyses. For true disagreements (e.g., one coder thought baby was walking and the other did not), the bulk coder’s data were retained in the final dataset.
We used ANOVAs for age and surfaces comparisons, with Bonferroni corrections for post hoc tests. We used Pearson correlations to test relations among continuous measures with Cook’s bivariate outliers removed.
Missing data were rare (<4% of the dataset). Infants’ feet were off camera in brief segments (median duration = 5.6 s) that accumulated to 0 to 6 min per visit.
Infants contributed 54.5 to 115 min of unconstrained time across the two visits, M = 95.7 min (Figure 2B). Time constrained varied widely, M = 18.3% of each hour (range = 0% to 72%), and differed by age F(2,35) = 3.5, p = .042, η^2^ = .18. Thirteen month-olds infants were constrained marginally more (M = 15.8 min/hr, SD = 8.6) than 23-month-olds (M = 7.9 min/hr, SD = 8.4), p = .051, but no other comparisons approached significance. Time constrained was stable across the two visits, r(34) = .416, p = .012 (Figure 5A).
Further analyses considered only the time infants were unconstrained. Based on time unconstrained, all but 4 of the 13 remaining measures were stable (Figure 5B-E). The 9 significant correlations (with Cook’s bivariate outliers removed in 4 analyses) ranged from .33 to .69, all ps < .05. We combined data across visits in further analyses and calculated frequencies as rates per hour unconstrained (denoted as rates/hr for brevity).
Boys and girls did not differ on any measures, all ts(35) < .67, ps > .09. So sex was combined in further analyses.
Every infant generated an immense number of M = 642 total transitions/hr. Figure 1B shows the five-minute segment with the most transitions for each infant, with rows ordered from the least to the most total transitions/hr (right column, range = 153 to 1,179). Total number of transitions was unrelated to infants’ age, r(34) = .06, p = .938 (see mix of symbols in left column). Even novice 13-month-old walkers generated immense numbers of transitions of each type. In fact, the two infants with the fewest and most total transitions were 23-month-olds (one watched television in both visits, and the other ran around the house in continual loops).
Babies transitioned from sitting or crawling into upright postures M = 24.9 times/hr (range = 3.2 to 67.4). Infants transitioned between standing and walking M = 410.6 times/hr (range = 135.3 to 637). They transitioned between surfaces M = 179.3 times/hr (range = 13.8 to 429.3); and they transitioned between places 27.3 times/hr (range = 0 to 84.1). Every baby exhibited every type of transition, except for two infants who did not transition between places. Four of 6 transitions were intercorrelated (Table 1, top left panel).
How did babies generate so many transitions of such variety? It is a mathematical certainty that more categories will produce more transitions, so we conservatively grouped each type of transition into the fewest conceptually meaningful categories. Thus, immense numbers of varied transitions were not an artifact of our annotation scheme. Similarly, it is a mathematical certainty that a large number of brief bouts will generate more transitions than a small number of extended bouts. But there is no certainty as to how ‘frequent’ and ‘brief’ are infants’ natural behaviors, and we conservatively defined bout endings. Open space to move does not necessarily limit the number of transitions between surfaces and places because babies could generate dozens of transitions via repeated visits.
Transitions to upright were common for three Infants’ sitting bouts were brief and frequent, babies preferred walking over crawling, and falls offered opportunities to get up.
As shown in Figure 6A, infants generally did not sit still. The median duration of a sitting bout was M = 17.3 s (SD = 13.0); 45.2% of sitting bouts lasted less than 10 s, and only 18.3% lasted more than 1 min (range = .07 s – 31.1 min). Indeed, whereas 50% of adult sitting bouts last more than 30 min (Johansson, Mathiassen, Rasmusse, & Hallman, 2020), only 8 babies ever sat continuously for more than 10 min (1 couch-potato baby sat for 31.1 min in visit 1 and 46.2 min in visit 2 while watching TV, and the other 7 infants never exceeded 21.7 min). However, because sitting was frequent (M = 29 times/hr, SD = 14.1) and 56% of infants had at least one sitting bout over 5 min, infants spent nearly half their time sitting (M = 25.2 min/hr, SD = 9.2). The number of sitting bouts—but not accumulated duration of sitting—was highly correlated with transitions to upright, r(34) = .91, p = .001, (Table 1, column 1), and falls, r(34) = .84, p = .001 (Table 1, column 5).
Although walking infants can crawl, and the 13-month-olds had recently learned to walk, all infants preferred walking over crawling. Bouts of crawling were brief and rare across age, months walking, and open space in the home, all ps > .05 (Table 1, rightmost columns), and when infants did crawl, they stood up almost immediately. Infants who fell more also displayed more crawling bouts and longer durations of crawling (Table 1, column 5). As shown in Figure 6B, the median duration of a crawling bout was M = 2.8 s (SD = 1.3); 97% of crawling bouts lasted less than 10 s, and none lasted more than 48.8 s. Infants crawled only M = 11.1 times/hr (SD = 9.2), so they accumulated less than 1 min of crawling time (M = 43.8 s/hr, SD =.64). The number of crawling bouts and accumulated time crawling were positively correlated with transitions to upright (Table 1, column 1).
Finally, transitions to upright were common because frequent falls resulted in frequent bouts of sitting and thereby gave infants frequent opportunities to stand up. All but 4 infants fell. Across age, infants fell M = 5.1 times/hr, and the number of infant falls was correlated with transitions to upright (Table 1, column 1). Frequency of falls varied by age, F(2,35) = 11.2, p < .001, η^2^ = .41. Thirteen-month-olds fell more often (M = 11.4, SD = 9.4) than 18- and 23-month-olds (Ms = 2.3 and 1.7, SDs = 1.9 and 1.8, respectively), ps <.002. Similarly, infants with fewer months walking fell more frequently than infants with more months walking, r(34) −.59, p < .001. Indeed, the newest walker had the most falls (37 falls/hr) and the most transitions to upright (67.4/hr). Of infants who fell, time sitting was shorter (M = 8.2 s) after the fall than if they sat down deliberately (M = 18.3), t(30) = −3.7, p < .001, Cohen’s d = 1.35, suggesting that falls were unwelcome interruptions in upright activity. Conversely, accumulated time standing was negatively correlated with transitions to upright (Table 1, column 1)
As in prior studies of infants in laboratory playrooms (e.g., Cole et al., 2016; Hoch et al., 2019), walking bouts were brief and frequent. However, we did not expect so many walking bouts to be so brief. As shown in Figure 6C, the average median duration of a walking bout was 2.1 s (SD = 0.6); 91% of walking bouts lasted less than 10 s and only 0.1% more than 1 min. Similarly, bouts of standing were brief and frequent. We conservatively defined the end of a walking bout as a stationary period of at least 1 s, such that the shortest standing period between walking bouts was 1 s. As shown in Figure 6D, the average median duration of a standing bout was 3.2 s (SD = 0.6); 87.5% of standing bouts lasted less than 10 s and only 0.3% lasted over 1 min. Thus, transitions between walking and standing were common because brief, frequent bursts of walking were separated by brief, frequent periods of standing. Moreover, we discovered that infants can transition from sitting to walking and from walking to sitting without an intermediate period of prolonged standing; 12% of standing bouts were less than 1 s because infants quickly transitioned between sitting and walking without a 1-s pause.
Notably, because walking and standing were so frequent (M = 206.2 walking bouts/hr, SD = 61.8 and M = 197.9 standing bouts/hr, SD = 61.7), more than half of infants’ time was upright. Infants spent more time standing, M = 19.4 min/hr (SD = 6.7), than walking, M = 14 mins/hr (SD = 5.5), t(35) = −4.3, p < .001, Cohen’s d = 1.45. Walking and standing behaviors did not differ by infant age, months walking, or open space to move, all ps > .05 (Table 1, rightmost columns).
Infants with more walking bouts and more accumulated time walking generated more transitions between standing and walking (Table 1, column 2). Infants who spent more time sitting and who had more bouts of crawling had fewer transitions between standing and walking (Table 1, column 2).
Infants averaged 179.3 transitions between surfaces by continually changing their location from step to step, most often walking back and forth between adjacent surfaces but occasionally taking long treks over floors covered in different surfaces. Notably, the primary living spaces in each home had both hard and soft flooring (no wall-to-wall carpet), non-floor surfaces (e.g., couch), and objects strewn on the floor (books, toys), thereby providing opportunities for transitions between surfaces. Every infant stepped on hard surfaces, all but 1 stepped on soft surfaces, all but 2 stepped on non-floor surfaces, and every infant stepped on objects.
Just as we found for other bouts of behavior, surface bouts were brief—only 2 s on average (range = 0.71 to 8.3 s). As expected, infants accumulated longer periods of time upright on hard (M = 17.8 min/hr) and soft surfaces (M = 12.9 min/hr) than on non-floor surfaces (M = 0.7 min/hr), or objects (M = 1.6 min/hr), F (1.63,57.18) = 74.77, p < .001, η^2^ = .68; post hoc ts(35) > 9.3, ps < .001. Moreover, infants frequently placed their feet on different types of surfaces simultaneously. The frequency of standing and stepping on mixed surfaces was greater (M = 103.1 times/hr, SD = 59.7) than for hard (M = 71.8 times/hr, SD = 42.5) and soft surfaces (M = 78.2 times/hr, SD = 58.9), non-floor surfaces (M = 4.7 times/hr, SD = 10.8), or objects (M = 3.9 times/hr, SD = 6.9), F (2.22,77.75) = 57.34, p < .001, η^2^ = .62; post hoc ts(35) > 3.7, ps < .001.
Infants who generated more walking and standing bouts and spent more time walking had more transitions between surfaces (Table 1, column 3), although total time standing was unrelated to surface transitions. Infants who spent more time sitting had fewer transitions between surfaces (Table 1, column 3). Transitions between surfaces did not differ based on infant age or months walking, all ps > .10, but infants with more open space to move generated more transitions between surfaces (Table 1, rightmost column).
Transitions between places necessarily underestimated the variety of paths for steering and navigation because we counted transitions based solely on rooms. Infants averaged 27 transitions between places largely by repeatedly visiting a few rooms. Although the number of infant-generated transitions walking between places was correlated with the number of different places infants went in visit 1, r(34) = .52, p < .002 and visit 2, r(34) = .37, p < .032, the number of transitions far exceeded the number of places visited for every infant (M = 4.2). Indeed, 50% of homes had 6 or fewer places, and mothers often constrained infants from going into specific places or carried infants from place to place. Infants accumulated the most time standing and walking in the common living areas of their home, M = 22.1 min/hr, followed by bedrooms M = 5.2 min/hr, and kitchens M = 3.0 min/hr, and the least time in hallways, M = 1.8 min/hr, and bathrooms, M = 0.1 min/hr.
As with surface transitions, infants who generated more walking and standing bouts and spent more time walking had more transitions between places (Table 1, column 4), although total time standing was unrelated to place transitions. Infants who spent more time sitting had fewer transitions between places (Table 1, column 4). As with surface transitions, place transitions were correlated with open space to move (Table 1, rightmost column).
We offer a novel conceptualization of what it means to practice a skill by focusing on naturally-occurring transitions in infant walking. Spontaneous transitions are a critical component of practice and a unique form of variability. Nonetheless, despite widespread agreement that postural transitions are foundational to functional behavior (Marsala & VanSant, 1998; Piper & Darrah, 2021), spontaneous transitions have received limited attention (c.f., Thurman & Corbetta, 2020). Moreover, transitions between states, behaviors, and contexts introduce varied opportunities for learning (Thelen & Smith, 1998; Thelen & Ulrich, 1991). Truly functional skills require self-generated, moment-to-moment adaptations to ever-changing affordances (Adolph & Hoch, 2019).
We focused on transitions to enter the task space, produce and pause the target skill, and use the skill in varied contexts. Both novice and experienced walkers—13–23 months of age—spontaneously produced Ms = 25 transitions from non-upright to upright, 411 transitions between standing and walking, 179 transitions between surfaces, and 27 transitions between places—totaling 642 transitions/hr. Notably, frequent transitions were not an artifact of our annotation scheme. We conservatively limited the number of categories for each type of transition, excluded transitions that facilitated others, and excluded transitions imposed by the mother. If anything, 642 transitions is an underestimate.
How did infants generate such immense numbers of transitions? Frequent transitions to upright and between standing and walking were inevitable because infants’ behaviors (sitting, crawling, standing, walking) were brief and frequent. To the extent that sitting and standing function as ‘rest periods,’ these breaks were typically brief. The newest walkers serendipitously had more opportunities to transition to upright because they fell more frequently than did older, more experienced walkers. Unexpectedly, transitions to upright did not necessitate an intermediate pause in standing. Infants sometimes transitioned from sitting or crawling to walking and vice versa without pausing for even 1 s.
More generally, infant behavior is naturally exuberant and flighty. Like frequent transitions in walking, babies flit between objects during play (Herzberg et al., 2022). Likewise, they vacillate in and out of social interactions with caregivers during free play or face-to-face interactions (Beebe et al., 2016; Karasik, Tamis-LeMonda, & Adolph, 2011; Suarez-Rivera, Pinheiro-Mehta, & Tamis-LeMonda, 2023). Unlike adults who sit for extended periods (Johansson et al., 2020), infants don’t sit still—at least without caregiver engagement in play (c.f., Hoch, Hospodar, Koch de Costa Aguiar Alves, & Adolph, 2024; Schatz, Suarez-Rivera, Kaplan, & Tamis-LeMonda, 2022). Even in structured tasks, infants’ visual attention fluctuates from target to target (Phillips, Goupil, Whitehorn, & Wass, 2023; Suarez-Rivera, Smith, & Yu, 2019), and bouts of focused manual exploration last only a few seconds (Ruff, 1986; Soska & Adolph, 2014).
Previous researchers interpreted infant exuberance as ‘distractibility,’ ‘lack of focus,’ and ‘immaturity’ of effortful control (Kannass, Oakes, & Shaddy, 2006; Power, Chapieski, & McGrath, 1985; Ruff & Capozzoli, 2003; Ruff & Lawson, 1990; Yarrow et al., 1983). To the contrary, we propose that infant exuberance is a critical feature of skill learning rather than a liability or a ‘bug’ in the system (Herzberg et al., 2022; D. K. Lee et al., 2018; Ossmy et al., 2018). Infant exuberance allows them to learn as much as possible about as many things as possible (Herzberg et al., 2022; Smith, Jayaraman, Clerkin, & Yu, 2018). Indeed, our findings suggest that effective interventions should introduce variability that encourages infants to produce moment-to-moment transitions in behavior. Practice taking continual, alternating steps on a motorized treadmill, for example, can strengthen infants’ legs (Angulo-Barroso, Wu, & Ulrich, 2008; Y. G. Han & Yun, 2020; Wu, Looper, Ulrich, Ulrich, & Angulo-Barroso, 2007), but it cannot teach infants to get their bodies upright and to start and stop walking at will. Self-generated variability provides infants with enormous practice entering and exiting the task space and producing and pausing the target behaviors.
Another reason for frequent transitions to upright and between standing and walking is that infants push their limits. They deliberately change states (e.g., between sitting and standing and between standing and walking), thereby shifting between contexts of equilibrium and disequilibrium that can facilitate learning (Thelen & Smith, 1998; Thelen & Ulrich, 1991). Rather than staying inside their ‘comfort zone,’ babies push their zone of proximal development (Vygotsky, 1930/1978)—even if new skills are more challenging, energetically costly, and prone to errors. Standing requires more work, postural control, and strength than sitting (Serra-Ano Lopez-Bueno, García-Masso, Pellicer-Chenoll, & Gonzalez, 2015). Walking is more energetically costly than standing and more prone to falling (Levine, Schleusner, & Jensen, 2000). Running, spinning, and jumping are even more costly and susceptible to errors. But infants spontaneously produce whatever behaviors they can immediately (or nearly) do.
Fortuitously, infants are self-motivated to practice new skills like walking that supplant prior solutions like crawling (Adolph et al., 2012; Adolph & Tamis-LeMonda, 2014; Franchak, Kretch, & Adolph, 2018). Across age and walking experience, infants rarely crawled (44 s/hr, on average)—and crawled only to access nearby objects or to climb onto furniture, not to travel long distances. Infants were eager to get upright from sitting and after falling—spending half their time upright. Indeed, pre-walking 12-month-olds spend more time standing upright than they do crawling (Franchak et al., 2018). Crawling is such a temporary, dispensable solution in development that many children skip crawling altogether and the US Center for Disease Control eliminated it as an obligatory, universal ‘milestone’ (CDC US Center for Disease Control, 2025; Kretch et al., 2022; Zubler et al., 2022).
Likewise, in other domains, such as language, infants are motivated to practice new skills, gradually producing conventional words that supplant prior solutions like babbling and gesture (McGillion et al., 2017) despite difficulties in pronunciation into early childhood (Cychosz, Munson, & Edwards, 2021). However, motivation to move is likely hindered in children with motor disabilities, and errors like falling can be extremely costly. Whereas typically-developing babies are back upright within seconds after a fall, children with motor disabilities can require several minutes and caregiver assistance to get upright again (Prosser, 2025). Whereas typically-developing infants rarely fuss and caregivers rarely show concern after babies fall (D. Han & Adolph, 2021), falls in children with motor disabilities are likely to incur injury (Esterley et al., 2024), so caregivers and clinicians rarely allow children to fall freely (Skorup, Pierce, Paremski, Alcott, & Prosser, 2024).
Infants generated 179 transitions between surfaces and 27 transitions between places, not because any home had 179 surfaces or 27 places, but because of repetitions and revisits. For example, the infant with the smallest home walked in only 1 of the 4 available places (Figure 3), but she still accumulated 210 surface transitions/hr by retracing her paths over carpet and wood flooring, walking on the couch and two chairs, and stepping on 12 objects (not shown). The infant in the largest home walked in 12 of the 16 available places, but still revisited old haunts, stepping on varied surfaces enroute. Nonetheless, open space to move was correlated with the number of transitions between surfaces and places. Given that most infants lived in relatively small NYC apartments, the number of surface and place transitions are likely underestimates had we sampled more broadly, including infants living in relatively large single-family homes.
Across infants, surface transitions were primarily between hard and soft surfaces, and often on both surfaces at once (Figure 4C). But babies also stepped on non-floor surfaces (couch, bed), and to our surprise, objects strewn on the floor that created uneven 3-dimensional terrain that sometimes shifted underfoot. The infant shown in Figure 4C, for example, stood with one foot straddling an object, carpet, and wood and the other foot on the handle of a cabinet. Infants may have stepped on objects because they did not view them as ‘obstacles’ to be avoided or because they could not navigate over or around them (Hospodar, Elasmar, Candelaria, Liu, & Adolph, 2024), but they likely saw the objects given their eyes are so close to the floor (Franchak, Kretch, Soska, & Adolph, 2011).
Infants averaged 27 transitions between places by revisiting a small number of places repeatedly. Even when infants stayed in one room, they still steered around furniture, accommodating their walking paths to the environmental layout (Figure 3). If we had counted caregiver-induced transitions between places, the number of place transitions would have been even higher.
As Bernstein (1996) put it, everyday practice regimens involve “repetition without repetition” (p. 204). Just as no step is exactly the same as the last step, revisits to surfaces and places are never exactly the same. Likewise, behaviors in other domains (talking, looking, emotional expressions, object interactions, social interactions, etc.) are simultaneously repetitive and varied (Smith et al., 2018). Such repetition without repetition is creative and generative—and leads to better functional adaptation to variations in local conditions and greater transfer of skills from one context to another.
Documenting infants’ experience is laborious, but necessary. To characterize the input for infant learning, researchers observe ‘naturalistic’ behavior in a controlled laboratory environment or ‘natural’ behavior in the uncontrolled home environment. Naturalistic lab studies have the same room layout and objects by design and infant-caregiver dyads are instructed “to play.” In contrast, each infant’s home layout, objects, and activities are unique, and we instructed mothers to “ignore us and go about their normal day.” Nonetheless, naturalistic observations reveal aspects of behavior that hold across any setting. Infants’ spontaneous walking—wherever it is observed—contains time-distributed bouts of varied durations, speeds, path shapes, and step directions, different fall outcomes, and so on (Adolph et al., 2012; D. Han & Adolph, 2021; Hoch et al., 2019; D. K. Lee et al., 2018). But only natural observations of spontaneous activity in the familiar home environment reveal the actual contexts and quantities of everyday behaviors—where and how much infants walk, stand, sit, crawl, fall, and the number of transitions between behaviors, surfaces, and places.
Compared to naturalistic lab studies of infant walking, where infants spend 33–50% of each hour walking (Adolph et al., 2012; Cole et al., 2016; Hoch et al., 2024; Hoch et al., 2019), infants at home walked less—only 23% of each unconstrained hour, on average. Why the difference? Of course, time constrained (strapped in highchairs and so on) is surely higher at home than in lab, but here we analyzed only unconstrained time, so time constrained did not account for the difference. Another possibility is that the novelty of a lab playroom filled with new toys and places instigates more locomotor exploration than the familiar objects at home. However, babies move just as much in an empty lab playroom as in a playroom filled with novel toys (Hoch et al., 2019). A third possibility—supported by our data—is that the natural home environment supports unconstrained yet sedentary activities that are unlikely in a lab playroom (screentime, cuddling or book reading with caregiver, eating snacks, etc.). Indeed, one 23-month-old watched TV with mother for 31 min, another read books with mother on the couch for 22 min, and so on.
Each baby generated their own signature of transitions. As in naturalistic lab studies (Adolph et al., 2012; Hoch et al., 2019; Thurman & Corbetta, 2017), infants at home displayed large individual differences in postural and locomotor behaviors (e.g., 153 to 1179 transitions/hr, Figure 1B) that were stable across visits. In addition to presumed differences in arousal level (Buss, Perez-Edgar, Vallorani, & Anaya, 2019), individual differences in the quantity and context of walking and transitions depend on the physical features of infants’ home environment and their everyday activities. Open space was positively correlated with transitions between surfaces and places (rs = .34 and .43, respectively); and time sitting was negatively correlated with transitions between standing and walking (r = −.92) and with transitions between surfaces and places (rs = −.41 and −.45, respectively). For example, the 23-month-old, TV-watching, ‘couch potato’ generated 87% of her 154 transitions by switching between standing and walking, and had only 5 transitions to upright, 14 transitions between surfaces, and only 1 transition between places. In contrast, the 23-month-old ‘whirling dervish’ at the bottom of Figure 1B generated 54% of his 1146 transitions by switching between standing and walking, and had 41 transitions to upright, 406 transitions between surfaces, and 82 transitions between places.
Functional walking—like learning any skill—requires infants to adapt their behaviors from moment to moment to ever-changing affordances. Transitions between behaviors and environmental contexts are a vital, but understudied, component of practice. Infants generate immense numbers of walking transitions during everyday activity. Likely, the practice regimen for learning to talk, manipulate objects, and engage in social interactions also comprise immense numbers of transitions to enter the task space, produce and pause the target skill, and use the behavior in varied contexts.