1. Introduction

Parkinson's disease has progressively affected postural and motor control in many, a prevalent issue today. While pharmacological therapy and physical rehabilitation offer symptomatic relief, they do not target the underlying sensorimotor deficits that make postural instability so dangerous: up to 70% of patients fall at least once annually, and over half experience repeated falls (Allen et al., 2013). There are many existing mobility aids, but they fail to address the root cause of the issue, instead providing temporary pain relief. This project explores a novel approach: a wearable assistive tail worn at the lower back region, using balancing mechanisms to assist with motor control. Inspired by the movement of animals, this project aims to provide users with something that patients can use naturally in daily life without clinical supervision.

2. Research Question

How can a wearable assistive tail effectively augment postural stability and spatial awareness in individuals with Parkinson's disease, as measured through quantitative balance and gait metrics? This is the basis of our research.

3. Preliminary Experiment

Before designing the tail, a preliminary observational experiment was conducted to identify the most common movement abnormalities in patients.  There was a pronounced tendency in patients toward external rotation of the knees and feet, producing a turned-out, duck-footed stance. This is a symptom of Parkinson's disease, which weakens the hip flexor and internal rotator muscles responsible for keeping the legs aligned during walking (Mak et al., 2017). The result is a laterally unstable base of support that forces patients into a shuffling, flat-footed gait with shorter steps and reduced cadence, making them unable to shift weight effectively. While patients may feel safer, it actually increases fall risk by limiting the patient's ability to take corrective balancing steps. This posture also shifts the center of mass, creating a chronic imbalance that infringes on motor control skills. Even an uneven surface or a slight incline can easily overwhelm their corrective balancing skills, making it easier for patients to lose their balance and fall. We tested patients’ balance and walking without weights, then attached weights to a tail without circuit. Through this process, we also noticed that they tended to lean forward and will use technology to correct posture.

4. Design and Methodology

4.1 Hardware

To accurately measure and process postural stability data and deliver mechanical counterbalancing responses, the wearable assistive tail is equipped with the following instrumentation (Figure 1):

This includes an ultrasonic sensor, force-sensitive resistors, haptic contact sensor, accelerometer, servo motor assembly, 3D-printed articulated tail segments, counterweight housing, microcontroller, battery, charging port, and wires. 

The ultrasonic sensor is mounted at the anterior side of the harness, oriented horizontally and parallel to the ground:

• measures the distance between the user and a fixed reference surface
• Captures movement in real time
• Sensor calibration takes place during the initialization phase, as outlined in 4.2.1

The force-sensitive resistors are embedded within a custom plantar insole worn by the participant on each foot

• detect foot pressure during walking
• enables computation of ground contact time per foot and the step imbalance index between sides
• FSR calibration also takes place during the initialization phase

The haptic feedback motors consist of two vibrotactile units embedded in the posterior section of back harness at left & right lower back positions.

• delivering directional vibration cues to the participant in response to perturbation classification
• reinforcing posture correction along with the tail's mechanical stabilization

The accelerometer is integrated into the harness:

• measure three-axis trunk acceleration at 100 Hz throughout each trial
• captures the speed and magnitude of perturbations in real time
• data is used to compute perturbation detection speed, defined as the time elapsed between perturbation onset and the moment peak acceleration is reached

Data collected by the above sensors will be transmitted to the microcontroller through a network of wires

• microcontroller is integrated into the harness
• process all input data, compute postural stability metrics, classify perturbation events, trigger the appropriate tail actuation response
• transmit all recorded data to an external computer device via Bluetooth communication.

The device's battery will provide the sensors, microcontroller, and servo assembly with power. The battery will be recharged through a designated charging port. The counterweight housing and tail segment assembly will be accessible by detaching the distal segment of the tail, along the dotted line in Fig. 1.

4.2 Software

As shown in Figure 2, code will be implemented to deliver adaptive postural stabilization and record quantitative stability data. This will be split into several components:

• initialization of the device
• real-time sway detection
• perturbation classification
• tail actuation and haptic feedback response
• postural stability logging.

After the initialization phase, an iterative process alternates between two consecutive states: passive (counterweight stabilization only) and active (perturbation detected, servo actuation and haptic feedback engaged).

4.2.1 Initialization of the Device

The initialization phase is carried out in two steps to calibrate the device's sensors and establish the participant's baseline postural profile. 1. calibration of the ultrasonic sensor and accelerometer

• The participant stands upright facing the reference surface for 30s
• ultrasonic distance readings are averaged to define the session baseline distance
• accelerometer readings are averaged across all three axes to define the zero-acceleration reference frame

These values serve as the reference points for all subsequent sway and acceleration calculations (See  4.2.2 and 4.2.5). 2. Calibration of force-sensitive resistors

• participant stands still, the FSRs are read to establish a resting pressure average, defining the strike and liftoff detection thresholds used throughout the trial
• repeat this step at the start of each session, as footwear changes may affect sensor readings.

4.2.2 Sway Range Measurement

Ultrasonic sensor captures distance readings at 20 Hz throughout each trial.

• Each reading compared against the baseline distance established in Section 4.2.1 to compute an instantaneous sway value
• positive values indicate forward displacement and negative values indicate backward displacement
• sway range is calculated as the difference between the max and min sway values recorded over the trial window, as illustrated in Figure 3b:

---

4.2.3 Perturbation Recovery Time

A perturbation event is defined as any sway deviation exceeding 3.0 cm from baseline, as shown in Figure 5.

• perturbation recovery time is calculated as the elapsed time between perturbation onset and the moment sway returns within the threshold boundary
• upon detection, microcontroller simultaneously triggers the tail actuation response described in Section 4.2.4 and the haptic feedback response, and begins logging the recovery interval

---

4.2.4 Ground Contact Time and Step Imbalance Index

Plantar insoles detect pressure through FSR threshold crossings established in Section 4.2.1.

• ground contact time per foot is computed as the elapsed time between these events, as illustrated in Figure 4
• values are used to compute the step imbalance index using the formula in Figure 6
• step imbalance index of 0% indicates perfect symmetry, while values exceeding 15% are considered clinically significant in Parkinson's disease populations

---

4.2.5 Perturbation Detection Speed

When a perturbation is detected by the ultrasonic sensor, the microcontroller begins logging the three-axis acceleration data from the accelerometer at 100 Hz.

• perturbation detection speed is defined as the elapsed time between perturbation onset and the moment the resultant acceleration magnitude reaches its peak value (See Fig. 5)
• shorter detection speed in the without-tail condition indicates a more abrupt, unmitigated perturbation response
• a longer detection speed in the with-tail condition indicates that the tail's mechanical counterbalancing is reducing the event’s abruptness and giving the participant more time to engage voluntarily with corrective response

---

5. Analysis — Updated

5.1 Overview

Analysis evaluates the wearable assistive tail by metrics without tail and with tail, further subdivided by three tail weight configurations (lightweight: 400g, standard: 800g, heavyweight: 1200g). Four elderly female participants each completed three standardized tasks: standing, walking, and perturbation trial (or balancing depending on condition) under each configuration. All five postural stability metrics were recorded and compared across conditions for each participant.

5.2 Participant Information

Table 1. Participant demographics.

Participant	Age	Mobility Aid	Notes
P1	54	None	Youngest participant; highest baseline mobility
P2	59	None	Mild postural instability observed at screening
P3	68	Compression knee bands	Bilateral knee instability; reduced medial-lateral stability
P4	91	Walking stick	Oldest participant; most pronounced postural deficits

P4 presented the most significant postural instability during baseline screening, consistent with her age and reliance on a walking stick. P3's compression knee bands were not removed during testing, which may not eliminate confounding effects on FSR readings.

5.3 Metric 1 — Sway Range

Table 2. Mean sway range (cm) across conditions.

Participant	No Tail	400g	800g	1200g	Best
P1 (54)	3.2	2.6	1.9	1.4	1200g
P2 (59)	3.8	2.1	2.8	2.4	400g
P3 (68)	4.6	3.8	2.0	2.4	800g
P4 (91)	6.4	5.2	4.2	4.8	800g
Mean	4.50	3.50	2.73	2.68	—

• P1 showed consistent linear improvement across all weights, with 1200g producing the greatest reduction of 56.3%
• P2's best result was unexpectedly at 400g (2.4 cm)
  • performance worsening at 800g before recovering slightly at 1200g
  • suggesting irregular sensitivity to counterweight mass, possibly related to her asymmetric baseline gait pattern
• P3 peaked at 800g (2.0 cm) with performance reversing at 1200g (2.4 cm), indicating over-correction relative to her body frame
• P4's sway worsened from 4.2 cm at 800g to 4.8 cm at 1200g
  • Observed low body mass being unable to support 1200g added weight

5.4 Metric 2 — Perturbation Recovery Time

Table 3. Mean perturbation recovery time (ms) across conditions.

Participant	No Tail	400g	800g	1200g	Best
P1 (54)	1,620	1,280	916	676	1200g
P2 (59)	1,868	1,496	1,240	1,380	800g
P3 (68)	2,160	1,820	1,264	1,490	800g
P4 (91)	2,748	1,980	2,140	2,580	400g
Mean	2,099	1,644	1,390	1,532	—

• P1 showed consistent improvement across all weights, with 1200g reducing recovery time by 58.3% to 676 ms
• P2 + P3 recovery time worsened between 800g and 1200g (1,240 ms → 1,380 ms) (1,264 ms → 1,490 ms)
  • heavier load introduced instability during the recovery phase rather than aiding it
• P4's best recovery time was at 400g (1,980 ms), with performance deteriorating at both 800g and 1200g
  • even moderate counterweight mass exceeds her capacity to manage additional inertial load during dynamic recovery
  • 1200g recovery time of 2,580 ms nearly matched her without-tail baseline of 2,748 ms
  • heaviest configuration clinically ineffective for her

5.5 Metric 3 — Step Imbalance Index

Table 4. Mean step imbalance index (%) across conditions.

Participant	No Tail	400g	800g	1200g	Best
P1 (54)	18.3	11.2	9.9	7.1	1200g
P2 (59)	22.2	14.6	16.4	13.8	1200g
P3 (68)	32.8	18.4	14.8	17.4	800g
P4 (91)	45.1	31.2	28.4	34.6	800g
Mean	29.6%	18.9%	17.4%	18.2%	—

• P1 crossed below the 15% clinical significance threshold at 400g (11.2%) and continued improving to 7.1% at 1200g
  • the best gait symmetry result recorded across the entire dataset
• P2 showed a non-linear pattern
  • 400g produced a strong improvement to 14.2% (below threshold)
  • 800g worsened slightly to 16.4% (above threshold)
  • 1200g achieved the best result of 13.8%
  • P2's gait response to counterweight is sensitive to incremental weight changes
• P3's index fell below the clinical threshold only at 800g (14.8%), with 1200g causing a reversal to 17.4%, returning her above the threshold
• P4 showed the largest absolute improvement at 800g (45.1% → 28.4%), but neither 800g nor 1200g brought her below the clinical threshold, and 1200g worsened her index to 34.6%
  • may require a fundamentally different intervention approach beyond counterweight adjustment alone

6. Conclusion

5.6 Metric 4 — Perturbation Detection Speed

Table 5. Mean perturbation detection speed (ms) across conditions.

Participant	No Tail	400g	800g	1200g	Best
P1 (54)	42	68	94	128	1200g
P2 (59)	38	72	112	98	800g
P3 (68)	36	60	82	74	800g
P4 (91)	31	62	54	44	400g
Mean	36.8	65.5	85.5	86.0	—

• P1 consistently improved, reaching 128 ms at 1200g
  • longest detection speed recorded, indicating the 1200g tail most effectively distributed perturbation forces across her body mass
• P2, similar to P3 peaked at 800g before dropping at 1200g
  • the heavier configuration introduced a rebound acceleration effect that partially counteracted the force distribution benefit
• P4's detection speed peaked at 400g (62 ms) and declined at both 800g (54 ms) and 1200g (44 ms)
  • only participant whose detection speed worsened with increasing weight beyond 400g
  • 

5.7 Summary of Results

Table 6. Best-performing configuration per participant per metric.

Metric	P1 (54)	P2 (59)	P3 (68)	P4 (91)
Sway Range	1200g	400g	800g	800g
Recovery Time	1200g	800g	800g	400g
Step Imbalance	1200g	1200g	800g	800g
Detection Speed	1200g	800g	800g	400g
Overall Best	1200g	800g	800g	400g

• The optimal counterweight configuration is strongly participant-dependent rather than universally scalable
• P1 benefited most from the heaviest configuration across all metrics, consistent with strength providing sufficient resistance to manage the 1200g load
• P2 and P3 both performed best at 800g across the majority of metrics, with heavier configurations producing reversals in performance
  • particularly notable in P2's non-linear gait response and P3's consistent over-correction pattern driven by her bilateral knee instability
• P4's overall best configuration was 400g, with both 800g and 1200g producing deterioration across most metrics
  • Perhaps due to low body mass and advanced age creating an upper threshold for beneficial counterweight loading well below that of the other participants.

Thus, clinical prescription of the wearable assistive tail should involve individualised weight calibration based on participant age, body mass, and mobility profile rather than a standardised configuration.

5.8 Challenges and Resolutions

• Servo signal interference w/ sensor readings
  • PWM signals from the servo motors introduced electrical excess into sensor readings
  • resolved by adding 100µF decoupling capacitors across the power rails & introducing 20 ms delay between servo actuation and sensor sampling in the firmware
• Accelerometer drift during extended trials
  • MPU-6050 accumulated gyroscopic drift over sessions longer than 5min, introducing cumulative error into reference frame
  • resolved by automatic re-zeroing of the accelerometer baseline at start of each individual trial
• Ultrasonic false positives during walking
  • arm movements occasionally crossed the detection path during walk trials, triggering false perturbation events
  • resolved by implementing a 3-reading moving average filter requiring 3 consecutive threshold crossings before a perturbation is flagged
• FSR threshold variability between sessions
• resting FSR baselines shifted between sessions due to footwear and insole compression changes
• resolved by enforcing mandatory re-calibration at the start of every session
• P3 compression band interference
  • P3's compression knee bands altered plantar pressure distribution, skewing FSR readings
  • Could not resolve and had to prioritize patient safety
• Haptic motor vibration coupling into accelerometer
• vibrotactile haptic motors introduced low-frequency vibration artefacts into the accelerometer readings when firing simultaneously with data collection
• resolved by introducing a 15 ms blanking window in the firmware where accelerometer readings are paused immediately following haptic motor activation.
• 3D printed joint binding under 1200g load
  • articulated tail joints exhibited binding during full-range deflection under the heaviest counterweight
  • resolved by increasing joint clearance tolerances to 0.5 mm and lining tubing channels
• AA battery voltage drop during sustained actuation
  • simultaneous actuation of all four servos caused voltage drops that reset the microcontroller
  • resolved by staggering servo actuation so no more than 2 servos fire simultaneously
• P4 sensor displacement during perturbation trials
  • P4's reduced subcutaneous tissue caused sensor shift during trials, producing inconsistent readings
  • resolved by securing all sensors to maintain consistent contact pressure

7. Conclusion

• prioritize empirical data collection with a larger and more diagnostically diverse participant cohort

• prioritize empirical data collection with a larger and more diagnostically diverse participant cohort

  • including male participants and those across a broader range of Parkinson's disease severity stages
• integrate with more advanced machine learning models to adapt to user preferences
  • develop app to enhance statistics and functionality

Works Cited

Allen, N. E., Schwarzel, A. K., & Canning, C. G. (2013). Recurrent falls in Parkinson's disease: A systematic review. Parkinson's Disease, 2013, Article 906274. https://doi.org/10.1155/2013/906274 Libby, T., Moore, T. Y., Chang-Siu, E., Li, D., Cohen, D. J., Jusufi, A., & Full, R. J. (2012). Tail-assisted pitch control in lizards, robots and dinosaurs. Nature, 481(7380), 181–184. https://doi.org/10.1038/nature10710 Mak, M. K. Y., Wong-Yu, I. S. K., Shen, X., & Chung, C. L. (2017). Long-term effects of exercise and physical therapy in people with Parkinson's disease. Nature Reviews Neurology, 13(11), 689–703. https://doi.org/10.1038/nrneurol.2017.128 Shumway-Cook, A., Brauer, S., & Woollacott, M. (2000). Predicting the probability for falls in community-dwelling older adults using the Timed Up & Go Test. Physical Therapy, 80(9), 896–903. https://doi.org/10.1093/ptj/80.9.896 Winter, D. A. (2009). Biomechanics and motor control of human movement (4th ed.). John Wiley & Sons. https://doi.org/10.1002/9780470549148

4. Design and Methodology

4.1 Hardware

To accurately measure and process postural stability data and deliver mechanical counterbalancing responses, the wearable assistive tail is equipped with the following instrumentation (Figure 1):

This includes an ultrasonic sensor, force-sensitive resistors, haptic contact sensor, accelerometer, servo motor assembly, 3D-printed articulated tail segments, counterweight housing, microcontroller, battery, charging port, and wires.

The ultrasonic sensor is mounted at the anterior side of the harness, oriented horizontally and parallel to the ground:

• measures the distance between the user and a fixed reference surface
• Captures movement in real time
• Sensor calibration takes place during the initialization phase, as outlined in 4.2.1

The force-sensitive resistors are embedded within a custom plantar insole worn by the participant on each foot

• detect foot pressure during walking
• enables computation of ground contact time per foot and the step imbalance index between sides
• FSR calibration also takes place during the initialization phase

The haptic feedback motors consist of two vibrotactile units embedded in the posterior section of back harness at left & right lower back positions.

• delivering directional vibration cues to the participant in response to perturbation classification
• reinforcing posture correction along with the tail's mechanical stabilization

The accelerometer is integrated into the harness:

• measure three-axis trunk acceleration at 100 Hz throughout each trial
• captures the speed and magnitude of perturbations in real time
• data is used to compute perturbation detection speed, defined as the time elapsed between perturbation onset and the moment peak acceleration is reached

Data collected by the above sensors will be transmitted to the microcontroller through a network of wires

• microcontroller is integrated into the harness
• process all input data, compute postural stability metrics, classify perturbation events, trigger the appropriate tail actuation response
• transmit all recorded data to an external computer device via Bluetooth communication.

The device's battery will provide the sensors, microcontroller, and servo assembly with power. The battery will be recharged through a designated charging port. The counterweight housing and tail segment assembly will be accessible by detaching the distal segment of the tail, along the dotted line in Fig. 1.

4.2 Software

As shown in Figure 2, code will be implemented to deliver adaptive postural stabilization and record quantitative stability data. This will be split into several components:

• initialization of the device
• real-time sway detection
• perturbation classification
• tail actuation and haptic feedback response
• postural stability logging.

After the initialization phase, an iterative process alternates between two consecutive states: passive (counterweight stabilization only) and active (perturbation detected, servo actuation and haptic feedback engaged).

4.2.1 Initialization of the Device

The initialization phase is carried out in two steps to calibrate the device's sensors and establish the participant's baseline postural profile. 1. calibration of the ultrasonic sensor and accelerometer

• The participant stands upright facing the reference surface for 30s
• ultrasonic distance readings are averaged to define the session baseline distance
• accelerometer readings are averaged across all three axes to define the zero-acceleration reference frame

These values serve as the reference points for all subsequent sway and acceleration calculations (See  4.2.2 and 4.2.5). 2. Calibration of force-sensitive resistors

• participant stands still, the FSRs are read to establish a resting pressure average, defining the strike and liftoff detection thresholds used throughout the trial
• repeat this step at the start of each session, as footwear changes may affect sensor readings.

4.2.2 Sway Range Measurement

Ultrasonic sensor captures distance readings at 20 Hz throughout each trial.

• Each reading compared against the baseline distance established in Section 4.2.1 to compute an instantaneous sway value
• positive values indicate forward displacement and negative values indicate backward displacement
• sway range is calculated as the difference between the max and min sway values recorded over the trial window, as illustrated in Figure 3b:

4.2.3 Perturbation Recovery Time

A perturbation event is defined as any sway deviation exceeding 3.0 cm from baseline, as shown in Figure 5.

• perturbation recovery time is calculated as the elapsed time between perturbation onset and the moment sway returns within the threshold boundary
• upon detection, microcontroller simultaneously triggers the tail actuation response described in Section 4.2.4 and the haptic feedback response, and begins logging the recovery interval

4.2.4 Ground Contact Time and Step Imbalance Index

Plantar insoles detect pressure through FSR threshold crossings established in Section 4.2.1.

• ground contact time per foot is computed as the elapsed time between these events, as illustrated in Figure 4
• values are used to compute the step imbalance index using the formula in Figure 6
• step imbalance index of 0% indicates perfect symmetry, while values exceeding 15% are considered clinically significant in Parkinson's disease populations

4.2.5 Perturbation Detection Speed

When a perturbation is detected by the ultrasonic sensor, the microcontroller begins logging the three-axis acceleration data from the accelerometer at 100 Hz.

• perturbation detection speed is defined as the elapsed time between perturbation onset and the moment the resultant acceleration magnitude reaches its peak value (See Fig. 5)
• shorter detection speed in the without-tail condition indicates a more abrupt, unmitigated perturbation response
• a longer detection speed in the with-tail condition indicates that the tail's mechanical counterbalancing is reducing the event’s abruptness and giving the participant more time to engage voluntarily with corrective response

5. Analysis

5.1 Overview

Analysis evaluates the wearable assistive tail by metrics without tail and with tail, further subdivided by three tail weight configurations (lightweight: 400g, standard: 800g, heavyweight: 1200g). Four elderly female participants each completed three standardized tasks: standing, walking, and perturbation trial (or balancing depending on condition) under each configuration. All five postural stability metrics were recorded and compared across conditions for each participant.

5.2 Participant Information

Table 1. Participant demographics.

Participant	Age	Mobility Aid	Notes
P1	54	None	Youngest participant; highest baseline mobility
P2	59	None	Mild postural instability observed at screening
P3	68	Compression knee bands	Bilateral knee instability; reduced medial-lateral stability
P4	91	Walking stick	Oldest participant; most pronounced postural deficits

P4 presented the most significant postural instability during baseline screening, consistent with her age and reliance on a walking stick. P3's compression knee bands were not removed during testing, which may not eliminate confounding effects on FSR readings.

5.3 Metric 1 — Sway Range

Table 2. Mean sway range (cm) across conditions.

Participant	No Tail	400g	800g	1200g	Best
P1 (54)	3.2	2.6	1.9	1.4	1200g
P2 (59)	3.8	2.1	2.8	2.4	400g
P3 (68)	4.6	3.8	2.0	2.4	800g
P4 (91)	6.4	5.2	4.2	4.8	800g
Mean	4.50	3.50	2.73	2.68	—

• P1 showed consistent linear improvement across all weights, with 1200g producing the greatest reduction of 56.3%
• P2's best result was unexpectedly at 400g (2.4 cm)
  • performance worsening at 800g before recovering slightly at 1200g
  • suggesting irregular sensitivity to counterweight mass, possibly related to her asymmetric baseline gait pattern
• P3 peaked at 800g (2.0 cm) with performance reversing at 1200g (2.4 cm), indicating over-correction relative to her body frame
• P4's sway worsened from 4.2 cm at 800g to 4.8 cm at 1200g
  • Observed low body mass being unable to support 1200g added weight

5.4 Metric 2 — Perturbation Recovery Time

Table 3. Mean perturbation recovery time (ms) across conditions.

Participant	No Tail	400g	800g	1200g	Best
P1 (54)	1,620	1,280	916	676	1200g
P2 (59)	1,868	1,496	1,240	1,380	800g
P3 (68)	2,160	1,820	1,264	1,490	800g
P4 (91)	2,748	1,980	2,140	2,580	400g
Mean	2,099	1,644	1,390	1,532	—

• P1 showed consistent improvement across all weights, with 1200g reducing recovery time by 58.3% to 676 ms
• P2 + P3 recovery time worsened between 800g and 1200g (1,240 ms → 1,380 ms) (1,264 ms → 1,490 ms)
  • heavier load introduced instability during the recovery phase rather than aiding it
• P4's best recovery time was at 400g (1,980 ms), with performance deteriorating at both 800g and 1200g
  • even moderate counterweight mass exceeds her capacity to manage additional inertial load during dynamic recovery
  • 1200g recovery time of 2,580 ms nearly matched her without-tail baseline of 2,748 ms
  • heaviest configuration clinically ineffective for her

5.5 Metric 3 — Step Imbalance Index

Table 4. Mean step imbalance index (%) across conditions.

Participant	No Tail	400g	800g	1200g	Best
P1 (54)	18.3	11.2	9.9	7.1	1200g
P2 (59)	22.2	14.6	16.4	13.8	1200g
P3 (68)	32.8	18.4	14.8	17.4	800g
P4 (91)	45.1	31.2	28.4	34.6	800g
Mean	29.6%	18.9%	17.4%	18.2%	—

• P1 crossed below the 15% clinical significance threshold at 400g (11.2%) and continued improving to 7.1% at 1200g
  • the best gait symmetry result recorded across the entire dataset
• P2 showed a non-linear pattern
  • 400g produced a strong improvement to 14.2% (below threshold)
  • 800g worsened slightly to 16.4% (above threshold)
  • 1200g achieved the best result of 13.8%
  • P2's gait response to counterweight is sensitive to incremental weight changes
• P3's index fell below the clinical threshold only at 800g (14.8%), with 1200g causing a reversal to 17.4%, returning her above the threshold
• P4 showed the largest absolute improvement at 800g (45.1% → 28.4%), but neither 800g nor 1200g brought her below the clinical threshold, and 1200g worsened her index to 34.6%
  • may require a fundamentally different intervention approach beyond counterweight adjustment alone

6. Further Analysis

6.1 Metric 4 — Perturbation Detection Speed

Table 5. Mean perturbation detection speed (ms) across conditions.

Participant	No Tail	400g	800g	1200g	Best
P1 (54)	42	68	94	128	1200g
P2 (59)	38	72	112	98	800g
P3 (68)	36	60	82	74	800g
P4 (91)	31	62	54	44	400g
Mean	36.8	65.5	85.5	86.0	—

• P1 consistently improved, reaching 128 ms at 1200g
  • longest detection speed recorded, indicating the 1200g tail most effectively distributed perturbation forces across her body mass
• P2, similar to P3 peaked at 800g before dropping at 1200g
  • the heavier configuration introduced a rebound acceleration effect that partially counteracted the force distribution benefit
• P4's detection speed peaked at 400g (62 ms) and declined at both 800g (54 ms) and 1200g (44 ms)
  • only participant whose detection speed worsened with increasing weight beyond 400g
  • 

6.2 Summary of Results

Table 6. Best-performing configuration per participant per metric.

Metric	P1 (54)	P2 (59)	P3 (68)	P4 (91)
Sway Range	1200g	400g	800g	800g
Recovery Time	1200g	800g	800g	400g
Step Imbalance	1200g	1200g	800g	800g
Detection Speed	1200g	800g	800g	400g
Overall Best	1200g	800g	800g	400g

• The optimal counterweight configuration is strongly participant-dependent rather than universally scalable
• P1 benefited most from the heaviest configuration across all metrics, consistent with strength providing sufficient resistance to manage the 1200g load
• P2 and P3 both performed best at 800g across the majority of metrics, with heavier configurations producing reversals in performance
  • particularly notable in P2's non-linear gait response and P3's consistent over-correction pattern driven by her bilateral knee instability
• P4's overall best configuration was 400g, with both 800g and 1200g producing deterioration across most metrics
  • Perhaps due to low body mass and advanced age creating an upper threshold for beneficial counterweight loading well below that of the other participants.

Thus, clinical prescription of the wearable assistive tail should involve individualised weight calibration based on participant age, body mass, and mobility profile rather than a standardised configuration.

6.3 Challenges and Resolutions

• Servo signal interference w/ sensor readings
  • PWM signals from the servo motors introduced electrical excess into sensor readings
  • resolved by adding 100µF decoupling capacitors across the power rails & introducing 20 ms delay between servo actuation and sensor sampling in the firmware
• Accelerometer drift during extended trials
  • MPU-6050 accumulated gyroscopic drift over sessions longer than 5min, introducing cumulative error into reference frame
  • resolved by automatic re-zeroing of the accelerometer baseline at start of each individual trial
• Ultrasonic false positives during walking
  • arm movements occasionally crossed the detection path during walk trials, triggering false perturbation events
  • resolved by implementing a 3-reading moving average filter requiring 3 consecutive threshold crossings before a perturbation is flagged
• FSR threshold variability between sessions
• resting FSR baselines shifted between sessions due to footwear and insole compression changes
• resolved by enforcing mandatory re-calibration at the start of every session
• P3 compression band interference
  • P3's compression knee bands altered plantar pressure distribution, skewing FSR readings
  • Could not resolve and had to prioritize patient safety
• Haptic motor vibration coupling into accelerometer
• vibrotactile haptic motors introduced low-frequency vibration artefacts into the accelerometer readings when firing simultaneously with data collection
• resolved by introducing a 15 ms blanking window in the firmware where accelerometer readings are paused immediately following haptic motor activation.
• 3D printed joint binding under 1200g load
  • articulated tail joints exhibited binding during full-range deflection under the heaviest counterweight
  • resolved by increasing joint clearance tolerances to 0.5 mm and lining tubing channels
• AA battery voltage drop during sustained actuation
  • simultaneous actuation of all four servos caused voltage drops that reset the microcontroller
  • resolved by staggering servo actuation so no more than 2 servos fire simultaneously
• P4 sensor displacement during perturbation trials
  • P4's reduced subcutaneous tissue caused sensor shift during trials, producing inconsistent readings
  • resolved by securing all sensors to maintain consistent contact pressure

7. Conclusion

• prioritize empirical data collection with a larger and more diagnostically diverse participant cohort
• prioritize empirical data collection with a larger and more diagnostically diverse participant cohort
  • including male participants and those across a broader range of Parkinson's disease severity stages
• integrate with more advanced machine learning models to adapt to user preferences
  • develop app to enhance statistics and functionality

Works Cited

Allen, N. E., Schwarzel, A. K., & Canning, C. G. (2013). Recurrent falls in Parkinson's disease: A systematic review. Parkinson's Disease, 2013, Article 906274. https://doi.org/10.1155/2013/906274 Libby, T., Moore, T. Y., Chang-Siu, E., Li, D., Cohen, D. J., Jusufi, A., & Full, R. J. (2012). Tail-assisted pitch control in lizards, robots and dinosaurs. Nature, 481(7380), 181–184. https://doi.org/10.1038/nature10710 Mak, M. K. Y., Wong-Yu, I. S. K., Shen, X., & Chung, C. L. (2017). Long-term effects of exercise and physical therapy in people with Parkinson's disease. Nature Reviews Neurology, 13(11), 689–703. https://doi.org/10.1038/nrneurol.2017.128 Shumway-Cook, A., Brauer, S., & Woollacott, M. (2000). Predicting the probability for falls in community-dwelling older adults using the Timed Up & Go Test. Physical Therapy, 80(9), 896–903. https://doi.org/10.1093/ptj/80.9.896 Winter, D. A. (2009). Biomechanics and motor control of human movement (4th ed.). John Wiley & Sons. https://doi.org/10.1002/9780470549148

Firstly, we would like to acknowledge Ms. Trainor for her guidance as our school supervisor for CYSF and for her support of this project. We would also like to thank Lu's Clinic for helping us in sourcing suitable volunteers for our experiment. We also want to acknowledge our parents for their moral support throughout the project and all the authors of the works we have referenced and learned from.