The Impact of Gait Pattern Personalization on the Perception of Rigid Robotic Guidance: A Pilot User Experience Evaluation

This pilot study reveals that while personalized gait patterns for exoskeletons can be accurately executed, they do not significantly improve user experience metrics like comfort or naturalness compared to standard patterns, as short-term adaptation to the robotic system appears to be the dominant factor influencing perception.

The Gist

Imagine you have a pair of high-tech robotic legs that can walk for you. These aren't just simple crutches; they are "exoskeletons" that physically grab your hips, knees, and pelvis and force them to move in a specific rhythm.

The big question this paper asks is: Does it feel better if the robot knows exactly how you walk, or is it okay if it just uses a "one-size-fits-all" walking pattern?

Here is the story of their experiment, explained simply.

The Setup: The "Dance Floor" Robot

The researchers built a super-advanced version of a walking robot (based on a medical device called the Lokomat). Think of it as a treadmill with a robotic suit strapped to your body.

• The Old Way: Usually, these robots use a "Standard" dance step. It's an average of how a thousand people walk. It's like a generic pop song that everyone knows.
• The New Way (Personalized): They created a special algorithm that looks at your height, weight, age, and speed to predict exactly how you would walk. It's like the robot composing a unique song just for your body.
• The Wild Card (Random): They also threw in a "Random" pattern, which was just a walking style picked from a stranger's database. This was the "off-key" song to see if people could tell the difference.

The Experiment: Ten Dancers

They invited ten healthy, young adults to put on the suit. They didn't tell them which pattern was which. Each person walked for three short rounds:

1. Walking to the Personalized beat.
2. Walking to the Standard (average) beat.
3. Walking to the Random beat.

After each round, they asked: "Did you enjoy this? Did it feel comfortable? Did it feel natural?" They also measured how hard the robot had to push against their legs.

The Big Surprise: The "New Shoe" Effect

The researchers expected the Personalized pattern to win hands down. They thought, "Of course, if the robot mimics your own unique walk, it will feel like magic!"

But that's not what happened.

1. The Patterns Didn't Matter Much: The participants couldn't really tell the difference between the "Personalized" song and the "Standard" song. Both felt about the same. Interestingly, the "Random" song felt a bit more jarring (the robot had to push harder against their legs), but even that wasn't a huge deal.
2. The Real Winner Was Time: The biggest factor wasn't what the robot did, but when they did it.
   • The first walk felt stiff, awkward, and unnatural.
   • The last walk felt significantly more comfortable and natural.

The Analogy:
Think of this like putting on a brand-new, stiff pair of leather boots.

• The Personalized Pattern is like having boots custom-made for your foot shape.
• The Standard Pattern is like buying off-the-shelf boots.
• The Result: Even if the boots are custom-made, if you've never worn them before, they still feel stiff and weird for the first 10 minutes. But after you've walked around for a while (the "adaptation" phase), your feet get used to the boots, and they start to feel great.

The study found that getting used to the robot (adaptation) was way more important than whether the robot was copying your specific walk.

Why Didn't the Personalization Work?

The researchers had a few theories:

• The Robot is Too Bossy: The robot was set to be very "stiff." It held the users' legs tightly, like a strict dance instructor. This rigidity might have drowned out the subtle differences between the walking patterns.
• Humans are Flexible: Healthy people's walking styles are actually quite similar to each other. There's a lot of natural wiggle room. So, the "average" walk and the "personalized" walk were probably closer than the researchers thought.
• The "Strap" Factor: The straps holding the robot to the legs were soft and squishy. This acted like a cushion, blurring the fine details of the robot's movements.

The Takeaway

If you are designing a robot to help people walk (especially for rehabilitation), don't obsess over making the robot's math perfect for every single person immediately.

Instead, focus on:

1. Getting the user comfortable with the machine first. The "adaptation" period is crucial.
2. Making the robot less rigid. If the robot is too bossy, it doesn't matter if the pattern is perfect; it will just feel uncomfortable.

In short: A custom-tailored suit feels amazing, but only after you've worn it for a while. If the suit is too stiff, you won't notice the tailoring until you've broken it in. The researchers learned that for these robots, breaking it in is the most important step.

Technical Summary

1. Problem Statement

Exoskeletons often enforce predefined kinematic trajectories to guide human movement for rehabilitation or performance augmentation. While rigid guidance is necessary for safety and injury prevention, it often conflicts with an individual's natural gait, which is inherently unique and shaped by anthropometry, demographics, and speed.

• The Gap: Although data-driven methods exist to personalize gait trajectories (predicting individual hip, knee, and pelvis paths), there is a lack of empirical evidence regarding whether this personalization actually improves the subjective user experience (comfort, naturalness, enjoyment) compared to standard or random patterns.
• The Question: Does the computational cost of generating personalized gait patterns justify the implementation over standard patterns, particularly in terms of user perception and interaction forces?

2. Methodology

A. Hardware: Modified Lokomat® Exoskeleton

The study utilized a custom-modified, grounded lower-limb exoskeleton based on the commercial Lokomat®.

• Degrees of Freedom (DoF): The system features 12 DoFs total, with 7 actuated DoFs:
  • Pelvis: 6 DoFs (translational and rotational), with one actuated DoF for lateral translation.
  • Hips: 2 DoFs per leg (flexion/extension and ab-/adduction) driven by a closed-chain mechanism.
  • Knees: 1 DoF per leg (flexion/extension).
• Control Strategy: The system operated in Proportional-Derivative (PD) position control with high stiffness ($K_p$ and $K_d$ tuned to minimize compliance). This "rigid" control was intentional to maximize the user's sensitivity to trajectory differences.
• Sensing: Included force sensors at the knees, IMUs, and motion cameras for pelvis tracking.

B. Gait Pattern Generation

Three distinct gait patterns were enforced:

1. Personalized Pattern: Generated using a data-driven regression framework.
   • Input Data: Walking speed, height, weight, age, and gender.
   • Training Data: A public database of 42 healthy adults (ages 21–84) walking at various speeds.
   • Model: Multiple polynomial regression models predicted key events (timing, position, velocity, acceleration) for hip, knee, and lateral pelvis trajectories. These events were interpolated using 5th-order quintic splines to create continuous trajectories.
2. Standard Pattern: An average of all gait patterns in the training database, adjusted for speed using a regression model for gait cycle time.
3. Random Pattern: A gait pattern randomly selected from the training database (unrelated to the participant's demographics).

C. Experimental Protocol

• Participants: 10 unimpaired adults (5 male, 5 female; mean age 25).
• Design: Within-subject design with three 2-minute trials (Personalized, Standard, Random). The order was pseudo-randomized (Latin square).
• Conditions: Participants walked on a treadmill at ~1.8 km/h (slow speed) while remaining passive. No body weight support was applied (harness was slack).
• Data Collection:
  • Subjective: Post-trial questionnaires rating Enjoyment (Intrinsic Motivation Inventory), Comfort (7 items), Naturalness (4 items), and Passiveness. Participants also ranked the patterns.
  • Objective: Mean Absolute Interaction Forces at the knee joints (100 Hz sampling).
• Analysis: Linear Mixed Models (LMM) were used to analyze subjective ratings and forces, accounting for pattern type and trial order. Friedman tests analyzed ranking data.

3. Key Contributions

1. Novel Hardware: Development and validation of a multi-planar exoskeleton with actuated hip ab-/adduction and lateral pelvis translation, allowing for more realistic gait simulation than standard knee-only devices.
2. Personalization Framework: Implementation of a regression-based prediction model that generates individualized trajectories for hip, knee, and pelvis based on limited anthropometric and demographic inputs.
3. Empirical Evaluation: A pilot study directly comparing the subjective and objective impacts of personalized vs. non-personalized rigid guidance, addressing a critical gap in Human-Robot Interaction (HRI) literature.

4. Results

A. Trajectory Accuracy

• Both Personalized and Standard patterns showed similar accuracy (RMSE) when compared to actual measured gait data from the training database.
• The Standard pattern was slightly more accurate for hip trajectories, while the Personalized pattern was slightly more accurate for knee and pelvis trajectories.

B. Subjective User Experience

• No Significant Difference by Pattern: There were no significant differences in Enjoyment, overall Comfort, or Naturalness between the Personalized, Standard, and Random patterns.
• Physical Strain: Participants reported slightly higher physical strain with the Personalized pattern compared to the Standard, though this did not translate to significant differences in interaction forces.
• Adaptation Effect (Significant): The order of trials had a dominant effect. The third trial was rated significantly as more comfortable and natural than the first trial, regardless of which pattern was enforced. This indicates rapid user adaptation to the exoskeleton.
• Rankings: Participants could not significantly distinguish between patterns when ranking them for preference, comfort, or naturalness.

C. Human-Robot Interaction Forces

• Random vs. Standard: The Random pattern generated significantly higher interaction forces at the knee joints compared to the Standard pattern (Right: $p=0.002$; Left: $p=0.001$).
• Personalized vs. Standard: Differences in interaction forces were not statistically significant, though the Personalized pattern showed a trend toward higher forces on the left leg ($p=0.061$).

5. Significance and Conclusions

• Minimal Short-Term Impact of Personalization: In the context of rigid, fully guided exoskeletons, personalizing gait kinematics based on anthropometry and speed has minimal short-term influence on user experience (comfort, naturalness, enjoyment) for unimpaired users.
• Dominance of Adaptation: The user's adaptation to the robotic system (ergonomics, interface, and control rigidity) is a stronger determinant of experience than the specific trajectory being enforced. Users adapted to the device within minutes, overshadowing subtle differences between gait patterns.
• Design Implications:
  • The "intrusiveness" of rigid guidance and the mechanical interface (straps, cuffs) may mask the benefits of personalization.
  • Future controllers should consider compliant control strategies or trajectory-free frameworks (e.g., impedance control) where personalization might manifest more clearly.
  • For patients with severe sensorimotor impairments, rigid guidance might be perceived as "stable," and personalization could yield different results than in unimpaired users.
• Conclusion: While personalization algorithms are technically feasible, their implementation in rigid exoskeletons requires careful consideration of user adaptation and interface compliance. The study suggests that for rigid guidance, the "standard" pattern may be sufficient for short-term user experience, and resources might be better spent on improving system ergonomics and control compliance.