Split-Belt Treadmill Training for Asymmetric Walking Patterns

Original Editor - Mason Trauger

Top Contributors - Malisha van der Berg, Mason Trauger, Vidya Acharya and Alexandra Stead  

Introduction

A split-belt treadmill is a specialised piece of gait training equipment used to study and rehabilitate asymmetric walking patterns. Unlike a standard treadmill with a single belt, it has two independently driven belts, allowing each limb’s speed to be controlled separately. This set-up allows clinicians and researchers to generate a controlled, sustained gait perturbation and observe how the nervous system recalibrates walking in response.

Split-belt treadmill

Utilisation of a split-belt treadmill can be used to address deficits in stride length, gait speed, weight-bearing, etc. The paradigm is most established in Stroke rehabilitation, but has also been applied in:

• Parkinson’s disease (PD).
• Multiple sclerosis (MS), spinal cord injury (SCI), cerebellar ataxia, and traumatic brain injury (TBI).
• Anterior Cruciate Ligament (ACL) reconstruction rehabilitation.
• Falls and gait instability in older adults.

The following video demonstrates the utilisation of a split-belt treadmill for stroke rehabilitation:

Mechanics

A typical split-belt protocol follows three phases:

1. Baseline (tied-belt): both belts move at the same speed.
2. Adaptation (split-belt): belts move at different speeds, commonly a 2:1 ratio, sustained for 5 to 15 minutes.
3. Washout/post-adaptation (tied-belt): belts return to equal speed, allowing measurement of aftereffects.

Aftereffects are a transient asymmetry in the opposite direction to the original perturbation. They reflect motor recalibration by the nervous system.

Some protocols insert brief “catch trials” (10 to 30 seconds of tied-belt walking during adaptation) to test whether the adapted pattern transfers to overground walking.

A representative research paradigm (used to test ACL reconstruction patients against controls) sequences six periods:

1. Overground baseline,
2. treadmill baseline (tied),
3. adaptation I (split, 2:1 ratio, ~12 min),
4. washout (tied),
5. a second adaptation period (adaptation II, to test retention),
6. and an overground post-adaptation period (to test transfer).

Types of perturbation

Belt behaviour can also be manipulated to create discrete perturbations rather than sustained asymmetry:

• Slip-like perturbations: one belt suddenly accelerates, most often at initial contact. This causes the foot to slide forward and destabilises balance backward.
• Trip-like perturbations: one belt suddenly decelerates or stops. This destabilises balance forward, simulating an obstacle-type disturbance.
• Mediolateral and tilt perturbations (pitch/roll/yaw): some systems can shift the platform sideways or tilt it, simulating slopes, cambers, or sudden changes of direction.

The following video shows the different types of perturbations:

Equipment and measurement

Systems Available

Commercial systems vary in capability:

• Grail and Caren Motek platforms allow anterior-posterior, mediolateral, and tilt perturbations.
• The Senly treadmill uses the Froude number, a ratio standardising perturbation intensity to the individual’s leg length, and to individualise perturbation intensity.
• Woodway systems built flush into the floor, speed adaptations; see following video on a Woodway system:

Measurement

Outcome measures typically include:

1. Step length asymmetry,
2. Step/Stance time asymmetry,
3. Double support ratio.

Derived indices are calculated from these to quantify adaptation magnitude, retention (savings), and overground transfer.

Notably, the broader gait-rehabilitation literature uses markedly inconsistent symmetry equations (symmetry index, ratio index, gait asymmetry index, symmetry angle, and various normalised variants), which can produce different results from the same underlying data; some evidence suggests the symmetry index (SI) is more sensitive than alternatives, but no consensus method exists.

Indications

The following conditions and populations have well documented evidential support:

• Stroke: the most extensively studied population for split-belt symmetry training.
• PD, including patients with freezing of gait (FOG).
• MS, SCI, TBI, and cerebral palsy (CP).
• Cerebellar ataxia: used primarily as a research model to isolate cerebellar contributions to gait adaptation.
• ACL reconstruction: to target persistent post-operative limb-loading asymmetry linked to the later development of post-traumatic osteoarthritis.
• Healthy older adults, particularly for fall-risk reduction and balance training.
• Knee osteoarthritis and lower-limb amputation, in perturbation-recovery research.
• Fall biomechanics research more broadly: controlled slip perturbations on split-belt systems allow systematic study of fall mechanics using precisely parameterised, unpredictable perturbations that are difficult to achieve with obstacle- or lubricant-based methods.

Physiological and Biomechanical Rationale

Two types of motor control

Split-belt walking engages two distinct control mechanisms:

1. Reactive parameters (e.g. stance time, stride length): change immediately upon exposure to the perturbation, don’t evolve further with continued exposure, and return to baseline instantly once removed, with no aftereffects. Governed largely at the spinal level via central pattern generators.
2. Adaptive parameters (e.g. step length asymmetry, double support ratio): change gradually over adaptation, produce aftereffects when the perturbation is removed, and reflect feedforward, predictive motor learning dependent on an intact cerebellum.

The forward model

The dominant explanatory model is that the cerebellum recalibrates an internal “forward model” of the body’s expected movement, driven by sensory prediction errors, which are mismatches between predicted and actual sensory feedback.

The following video shows split-belt treadmill walking one foot in reverse and the other forward, explaining the "forward model"

This is thought to occur via a cortico-cerebellar loop:

1. The cerebellum receives sensory input and an efference copy of the motor command
2. Purkinje cells integrate error signals to update the forward model
3. The updated model is relayed via the thalamus to the supplementary motor area and primary motor cortex
4. The motor command is refined accordingly

Evidence for cerebellar involvement

The following support cerebellar involvement in split-belt adaptation:

• Patients with cerebellar ataxia show intact reactive adjustments but impaired adaptation and absent aftereffects.
• Cerebellar transcranial direct current stimulation can modulate adaptation rate.
• fMRI studies show people with PD recruit the cerebellar locomotor region more than healthy controls during gait, suggesting a compensatory role as basal ganglia circuits become impaired.
• In stroke, an intact cerebellum is considered essential for patients to benefit from split-belt adaptation, and adaptation, though typically present, occurs at a reduced rate compared to healthy controls.
• In ACL-reconstructed patients, adaptation capacity for step length symmetry appears preserved despite disrupted joint afferent feedback, suggesting the CNS can compensate using other available sensory sources.

The energy minimisation hypothesis

An alternative (or complementary) account holds that split-belt adaptation is driven by minimising metabolic energy cost rather than purely correcting a symmetry “error”.

Mechanically, a split-belt treadmill can perform net positive work on a walker who adopts positive step length asymmetry (SLA), longer steps on the fast belt, by absorbing more braking force on the fast side and generating more propulsion on the slow side.

Contraindications and Precautions

The following contraindications and precautions apply:

• Cerebellar lesions/dysfunction: Since adaptation is cerebellum-dependent, patients with cerebellar damage may show intact reactive responses but little or no adaptive learning, limiting expected benefit. Stroke trials commonly exclude patients with cerebellar involvement for this reason.
• Older adults and neurologically impaired populations: these groups show less efficient, less economical recovery strategies (longer recovery steps, greater destabilisation, increased muscular co-contraction) in response to perturbations, and require closer supervision, harnessing, or spotting during training.
• Musculoskeletal impairments compromising gait, and comorbid neurological diagnoses beyond the primary condition being treated, have been used as exclusion criteria in clinical trials and warrant caution in practice.
• Slip-like versus trip-like perturbations carry different risk profiles:
  • Slip-like perturbations destabilise balance more rapidly (greater reduction in margin of stability)
  • Trip-like perturbations, though slower in onset, tend to produce larger overall balance disruption
  • Both require appropriate safety precautions during training
• Perturbation intensity must be carefully titrated: Excessively low intensities may fail to sufficiently challenge balance or elicit meaningful compensatory responses, while excessively high intensities can overwhelm recovery mechanisms, provoke fear of falling, and reduce confidence, motor performance, and postural control.
• Limited ecological validity: Even precisely controlled split-belt and tilting perturbations do not fully replicate the complexity and unpredictability of real-world walking surfaces, so clinical judgement is needed when translating treadmill-based gains to overground and community mobility goals.
• Functional benefit of symmetry restoration is not guaranteed: Improving spatiotemporal gait symmetry may have limited functional benefit on its own for some individuals post-stroke, so symmetry should not necessarily be treated as an end goal independent of broader functional outcomes.

Summary

Split-belt treadmill training uses a treadmill with two independently controlled belts to create gait asymmetry for rehabilitation and research. Patients walk through baseline, adaptation, and washout phases, with recalibration measured through “aftereffects” once belt speeds equalise again. It is most established in stroke rehabilitation, but also used in PD, SCI, CP, ACL rehabilitation, and older-adult fall-risk training.

It relies on cerebellum-dependent adaptive motor learning, thought to work via an internal predictive “forward model” updated by sensory prediction errors, though energy-minimisation may also contribute. Precautions include reduced benefit with cerebellar damage, need for supervision in vulnerable populations,, careful intensity titration, and limited real-world functional transfer.

Resources

Some systems available for research:

Motek

Treadmetrix

Bertec FIT5

Woodway