Walking Isn’t Always Rhythmic: Slow Gait Switches to a Discrete Control Mode, Study Finds

Walking feels like the most automatic rhythm the human body produces, a metronome of legs that ticks along without conscious effort. But a new study suggests that this rhythm is not a fixed property of gait. Instead, the brain appears to toggle between two fundamentally different control modes depending on how fast we walk, and the switch happens at a measurable threshold. At very slow speeds, walking breaks apart into a chain of discrete, step-like movements; at faster speeds, it settles into a smooth, self-sustaining oscillation. The findings, published in the journal iScience by Giulia Panconi, Diego Minciacchi, Riccardo Bravi, and Nadia Dominici of Vrije Universiteit Amsterdam and their collaborators, provide some of the strongest evidence yet that the discrete-versus-rhythmic framework developed for arm movements applies to locomotion as well.

The research builds on a long-standing theory in motor neuroscience that the central nervous system generates movement from a small set of building blocks known as motor primitives. Within this framework, discrete movements are modeled as point-attractor dynamics: the limb moves from one stable posture to another, then stops. Rhythmic movements, by contrast, are modeled as limit-cycle dynamics, producing continuous periodic oscillations with stable phase relationships. Decades of evidence, from neuroimaging to motor-learning experiments, indicate that these two movement classes recruit partly distinct neural networks. Discrete actions engage broader bilateral and non-primary motor areas, while rhythmic movements rely on more circumscribed cortical and cerebellar circuitry. Stroke patients, tellingly, often retain rhythmic arm movements better than discrete ones.

Most of that evidence, however, comes from the upper limb, where reaching is discrete by default and slow rhythmic movements have been shown to fragment into sequences of discrete submovements. Walking presents the mirror-image case: it is prototypically rhythmic, so finding discrete-like organization in gait is especially informative precisely where the framework is least expected. Slow walking is also clinically important, since reduced gait speed is common in aging and neurological disease, exactly the conditions in which rhythmic locomotor organization may be compromised. If walking speed acts as a control parameter that triggers phase-transition-like shifts in coordination, the transitions might even display hysteresis, the history dependence familiar from the classic walk-to-run transition, which occurs at a higher speed when accelerating than when decelerating.

To capture these shifts, the team recorded 18 healthy adults walking on an instrumented treadmill while a ten-camera motion capture system tracked reflective markers and wireless electrodes recorded activity from 15 lower-limb muscles on each side. Participants completed two protocols: in one, treadmill speed increased from 0.5 to 5.0 kilometers per hour in 0.5 km/h steps; in the other, it decreased along the same range. At each speed, participants completed 15 consecutive strides. The researchers then quantified movement smoothness using two complementary metrics, log dimensionless jerk (LDJ) in the time domain and spectral arc length (SPARC) in the frequency domain, applied to the antero-posterior displacement between the right and left ankles, a low-dimensional descriptor of the inter-limb oscillation that defines gait.

The smoothness results were striking. Both metrics became progressively less negative, indicating smoother movement, as speed increased. Mean LDJ values rose from roughly −16.0 at 0.5 km/h to about −10.5 at 5 km/h, while SPARC climbed from around −8.3 to −4.0. Crucially, the consistency of these changes across participants varied systematically with speed. Between 0.5 and 2.5 km/h, pairwise comparisons of smoothness showed high agreement, with 89 to 100 percent of participants showing significant differences between adjacent low-speed segments. Around 3.0 to 3.5 km/h, that agreement dropped to a moderate range, and at the highest speeds it collapsed almost entirely, with few or no participants showing significant differences between neighboring fast segments. The researchers interpret this intermediate band as a transition window in which individuals shift into a stable rhythmic attractor at slightly different speeds before converging at faster gait.

Smoothness alone, however, is an end-effector readout and cannot by itself prove that a kinematic regime shift reflects reorganized neural control rather than the mere biomechanical consequences of moving slowly. To resolve that ambiguity, the team turned to muscle synergy analysis, decomposing electromyographic signals with non-negative matrix factorization into time-invariant muscle weightings and time-varying activation profiles, selecting the minimum number of synergies that accounted for at least 90 percent of the variance. The number of modules required rose systematically with speed: at the slowest speeds, two synergies sufficed; at intermediate speeds, three; and from roughly 3 to 4 km/h onward, four synergies were needed, stabilizing at the fastest gait. In the decremental condition, the richer four-synergy organization persisted down to about 3 km/h before collapsing back to two at the slowest speeds.

That asymmetry between the incremental and decremental protocols is the study’s most intriguing detail. The neuromuscular system held on to its higher-dimensional coordination state longer when speed was falling than it took to build it when speed was rising, a pattern consistent with hysteresis in a dynamical system settling into stable attractors whose stability depends on a control parameter. The behavior echoes the well-documented hysteresis of the walk-run transition and suggests that the nervous system’s current coordination state depends on its recent history. Notably, smoothness itself showed only minimal directional differences, meaning the history dependence was evident primarily in the neuromuscular measure, while both measures converged on the same overall speed-dependent reorganization.

The link between the two levels of description was remarkably tight. Within every single participant, smoothness metrics and synergy number were strongly and positively correlated across speed segments, with a median Spearman correlation of 0.92 in both ramp directions. As gait became smoother with increasing speed, the modular organization of muscle activity grew richer, in every individual regardless of ramp direction. Merging analyses reinforced the picture: when stepping from a faster to an adjacent slower speed, the lower-speed synergies could be reconstructed with cosine similarities near 0.99 as non-negative combinations of the faster-speed modules, indicating that modules merge rather than vanish as speed drops. The authors draw a parallel to development and stroke recovery, where immature or impaired gait similarly shows fewer, merged modules, and they attribute the richer fast-walking organization to greater dynamic demands: stronger propulsion, tighter phase-specific timing, and increased balance and foot-clearance requirements.

The authors are careful about interpretation. Muscle synergies are estimated from data, and a successful reconstruction does not prove the recovered modules are neurally encoded primitives; the team treats them as a description of the structure of recorded muscle activity rather than direct evidence of discrete neural modules. Yet the speed-dependent changes were reproduced across two protocols, two decomposition levels, and an independently quantified kinematic transition, none of which would be expected from the factorization procedure alone. Limitations include treadmill rather than overground walking, some excluded EMG channels, and a predominantly female sample, though the within-subject design controls for stable participant characteristics.

The broader implications reach toward rehabilitation and wearable technology. Because movement smoothness can be estimated from inertial sensors, the discrete-rhythmic balance could eventually be monitored in real time during adaptive locomotion training, helping clinicians distinguish genuine control adaptations from the kinematic consequences of slow movement. And conceptually, the study extends the dynamic-primitives framework to its least expected domain, showing that even the body’s most iconic rhythm is, at its slowest, something else entirely: a sequence of discrete steps stitched together by a nervous system that switches control modes as the world demands.

Subject of Research: Speed-dependent transitions between discrete and rhythmic control regimes in human walking, assessed via kinematic smoothness and muscle synergies

Article Title: Speed-driven transitions between discrete and rhythmic dynamics in walking revealed by kinematic smoothness and muscle synergies

Article References: Panconi, G., Minciacchi, D., Bravi, R., & Dominici, N. (2026). Speed-driven transitions between discrete and rhythmic dynamics in walking revealed by kinematic smoothness and muscle synergies. iScience, 29(11), Article 117779. https://doi.org/10.1016/j.isci.2026.117779

Image Credits: AI Generated

DOI: 10.1016/j.isci.2026.117779

Keywords: human walking, motor control, muscle synergies, movement smoothness, motor primitives, gait analysis, hysteresis, electromyography, locomotion, dynamic primitives, treadmill, neuromuscular control