This paper helps answer a very practical question for my project:

Can I realistically get continuous heart / pulse data from a wrist wearable while somebody is moving through an airport?

The answer from this study is probably yes, but not perfectly.

That matters because I am considering a wrist-based sensing route, and one of my concerns has been whether movement would make PPG-derived pulse information too unreliable to be useful in a real airport journey.

This paper does not test airports, autism or sensory overload. What it does test very well is whether a wrist-worn device can estimate pulse rate continuously under both still and moving conditions.

What the paper actually tested

Chen et al. evaluated the Empatica EmbracePlus wristband across three independent clinical studies.

The device was compared against FDA-cleared ECG reference systems while participants completed supervised activities intended to represent real-life conditions.

Across the combined studies:

• 157 participants were included;
• 16,915 paired one-minute observations were analysed;
• 5,621 minutes were recorded under no-motion conditions;
• 11,294 minutes were recorded during motion;
• participants included a range of ages, body sizes, skin tones and health conditions.

This makes the paper much stronger for the narrow question of whether wrist PPG can still produce useful pulse-rate estimates during movement than many small proof-of-concept studies.

The most important technical point: it did not use PPG alone

This is probably the most relevant part for my design.

The EmbracePlus did not simply read a raw PPG signal and treat it as reliable.

Its pulse-rate algorithm used:

multiwavelength PPG + 3-axis accelerometer data + machine-learning processing

The accelerometer helped the algorithm identify and clean motion artefacts from the optical pulse signal.

Design implication

If I use wrist PPG in an airport, movement sensing should not sit beside the heart-rate signal as an unrelated extra. It can help the system understand when the optical signal itself may be being disturbed by movement.

This gives accelerometry two possible roles in my project:

1. signal-quality / artefact interpretation — helping make pulse estimates more robust;
2. context interpretation — helping distinguish a physiological change while rushing from one occurring while relatively still.

That makes accelerometry increasingly difficult to justify leaving out.

Motion reduced accuracy, but did not make the signal unusable

The device performed better when participants were still.

Under no-motion conditions, the reported ARMS error was 1.67 bpm.

Under motion conditions, it increased to 4.39 bpm.

So movement clearly matters.

However, the motion result still remained within the study's predefined clinical acceptance threshold of 5 bpm.

The mean absolute error was:

• 0.6 bpm without motion;
• 1.77 bpm during motion.

The authors also reported very strong agreement with ECG across both conditions.

For my project, I do not need clinical beat-to-beat precision. I am much more interested in whether pulse information can contribute to identifying changes and trends alongside movement and context.

This paper suggests that wrist-based pulse sensing is plausible for that purpose, provided I design around its limitations rather than pretending the measurement is perfect.

The signal can fail or become less reliable

The paper is also useful because it makes failure visible.

Of the available EmbracePlus data, some minutes were rejected because the pulse-rate algorithm considered the signal unreliable. The paper notes that this was typically associated with the device being worn too loosely.

Overall data availability was 98.17%, which is high, but not 100%.

There were also larger outliers during activities involving stronger forearm contractions. The male motion subgroup slightly exceeded the primary ARMS threshold, which the authors linked to stronger forearm movement and contractions.

That matters for an airport context because people will be:

• carrying or pulling luggage;
• lifting bags;
• gripping handles;
• using phones;
• opening documents;
• walking quickly;
• moving through security.

So I should assume there will be moments when the pulse estimate is less trustworthy.

Important caution

A wearable being accurate on average does not mean every reading is accurate. The system needs to tolerate missing, noisy or temporarily unreliable data.

Pulse rate is not the same as interpreting state

This paper validates pulse-rate estimation.

It does not validate what a pulse-rate change means.

That distinction is essential for my project.

A reliable pulse-rate increase could still be caused by:

• walking;
• rushing;
• lifting luggage;
• caffeine;
• excitement;
• illness;
• frustration;
• cognitive effort;
• or many other factors.

So this paper strengthens the measurement layer of my system, not the interpretation layer.

It supports:

“the wrist device can probably measure pulse reasonably well while the person is moving.”

It does not support:

“therefore the system knows the traveller is overloaded.”

Why trend analysis is particularly relevant

The intended use discussed in the paper focuses on continuous monitoring, retrospective review and trend analysis, rather than requiring perfect beat-to-beat accuracy at every instant.

That feels closer to the role I may need for biosensing.

My system may not need to react to one isolated heart-rate reading. It may be more defensible to look for a change over time relative to the person's baseline, while also considering movement and airport context.

For example:

pulse rises + accelerometer shows rapid walking

→ likely physical activity is contributing strongly.

pulse remains elevated relative to baseline + movement drops + queue becomes crowded / noisy

→ the system may have more reason to consider a check-in.

Even then, the conclusion is still “maybe ask”, not “we know the state.”

What it DOES establish

For my project, I think this paper provides useful evidence that:

• wrist-worn PPG can provide reasonably accurate continuous pulse-rate estimates during a range of everyday movements;
• motion makes PPG harder, but does not necessarily make it unusable;
• accelerometer information can be used directly to help deal with motion artefacts;
• a wrist device can combine sensing and algorithmic processing rather than treating each sensor independently;
• and continuous pulse trends from a wrist wearable are technically plausible for the type of mobile context I am designing around.

What it does NOT establish

This study does not show:

• that pulse rate detects sensory overload;
• that heart-rate change identifies stress or anxiety;
• that the same accuracy would be achieved in an unsupervised airport journey;
• that the device can always produce a reliable reading during hand/arm activity;
• that pulse rate alone is enough for my sensing architecture;
• or that a detected physiological change should automatically trigger support.

It is also important that this is a manufacturer-linked validation study. All authors disclosed financial relationships with Empatica, the company that manufactures EmbracePlus. The paper is still useful, particularly because the validation is detailed and multi-site, but I should keep that relationship visible when weighing the evidence.

What this means for my wearable decision

This paper makes the wrist route more credible.

I was previously comparing ring and wrist-based forms partly at the level of comfort and available sensors. This study adds another consideration: a research-grade wrist device can combine multiwavelength PPG and accelerometry and still obtain useful pulse information during movement.

The wrist option also remains attractive because it may support other signals I am considering, particularly EDA, depending on the final sensor architecture.

However, the result does not automatically mean that an Empatica-style wristband is the final form I should design.

I still need to ask:

• is the wrist comfortable and acceptable for autistic users?
• does it interfere with clothing or security processes?
• how tight does it need to be for reliable sensing?
• would a ring be less burdensome?
• and which signal set do I actually need before choosing the form?

What changed in my thinking

Before this paper, I was treating motion mainly as a confound that makes physiological sensing less reliable.

I now see a more nuanced picture.

Motion is still a problem, but it can also be measured and used by the algorithm to improve the signal.

The key takeaway for me is:

Wrist pulse sensing during movement looks technically plausible, but only because the system accounts for motion rather than ignoring it.

This strengthens my current working architecture:

personal baseline + pulse / physiological change + accelerometry + airport context → cautious interpretation → transparent check-in → traveller decides

It also gives me more confidence that a wrist wearable is worth continuing to explore alongside the ring rather than rejecting wrist PPG simply because the airport involves constant movement.

Source

Chen, W., Cordero, R., Lever Taylor, J., Pangallo, D.R., Picard, R.W., Cruz, M. and Regalia, G. (2024). Multicenter Evaluation of Machine-Learning Continuous Pulse Rate Algorithm on Wrist-Worn Device. Digital Biomarkers, 8(1), pp. 218–228.

Key points from my reading: Empatica EmbracePlus; multiwavelength PPG plus 3-axis accelerometry; machine-learning motion-artifact cleaning; 157 participants across three independent clinical studies; 16,915 paired one-minute observations; ARMS 1.67 bpm without motion and 4.39 bpm during motion; 98.17% pulse-rate data availability; isolated motion-related outliers; useful evidence for continuous pulse trend measurement, not evidence of internal-state detection.