Biofeedback involves using electrical instruments to measure a person’s biometric responses, including brainwaves, heart rate, skin conductance, facial expressions, respiration, peripheral skin temperature, and muscle tone [
1,
2]. These biometric signals are often referred to as physiological [
3,
4,
5] and psychophysiological signals [
6]. The applications of biofeedback are broad and include medical purposes, such as physical and occupational therapy [
7,
8,
9,
10], psychological clinics [
11,
12,
13,
14], and cognitive research [
15,
16,
17,
18]. To ensure the accuracy and quality of their work, researchers often use medical-grade devices in biofeedback studies. The results obtained from medical-grade equipment provide a comparable standard for other research, which is particularly important for medical applications. Therefore, the use of medically certified instruments in therapeutic studies is mandatory. However, the high price of such equipment can be a significant barrier to its widespread adoption. Manufacturers must acquire an FDA class II or CE IIa certification for their products to meet medical certification requirements, making them excellent and reliable, but costly.
1.1. Consumer-Grade Physiological Sensors Applied in Research
Cost can be a concern when using medically certified devices for non-medical purposes [
19]. In recent years, consumer-grade biometric devices have been extended to non-medical fields, such as measuring pupils’ attention in education [
20], brain–computer interfaces in human–computer interaction [
21], workers’ mental load in human–robot collaboration [
22], and affective computing [
23,
24]. Some consumer-grade biometric devices have been reported to be as accurate as medical-grade products [
25,
26] or can be used for health care purposes [
27].
When selecting wearable sensors for Industry 5.0 applications, the two top features to consider are system development kit (SDK) support for real-time data streaming and wireless communication protocols. Three communication protocols are widely used in the industry sector. Experts recognize that the Bluetooth Low Energy (BLE) protocol is much more important than ANT+ and Wi-Fi. Every wearable device on the market supports BLE, the mainstream communication protocol. Other features, such as weight and comfort for workers without obstructions, are also important considerations [
28].
Photoplethysmography (PPG) is prevalent in the application of individual health care [
27]. Bolanos et al. [
29] indicated that heart rate variability (HRV) derived from PPG has excellent potential to replace the one from ECG. Recently, no significant differences have been reported in the HRV features derived from PPG and ECG signals in time, frequency, and non-linear domains [
30]. However, PPG should be used only when the user is resting [
31,
32] due to motion artifacts caused by the movement of the PPG sensor over the tissue, skin deformation due to muscle contraction, and blood flow dynamics [
33,
34]. Nevertheless, an algorithm can reduce movement artifacts using user kinematics data from embedded accelerometers [
35]. PPG is a non-invasive, low-cost, and wearable wireless device that can be an alternative to electrocardiogram (ECG) technology for heart rate (HR) monitoring. Although ECG has been continuously improved in terms of measurement accuracy and wearing comfort, the flexibility, portability, and convenience for users have not been enhanced [
27]. In contrast, PPG does not require several electrodes to be placed on specific body locations. Users can wear a PPG device like a watch, with flexibility of movement, portability to any location, and convenience to monitor HR all day. These features make PPG suitable for non-medical activities [
27], such as user experience studies [
36]. In recent years, there has been an extension of the use of PPG to non-medical fields, such as measuring pupils’ attention in education [
20], brain–computer interfaces in human–computer interaction [
21,
22], and affective computing [
23,
24]. Some consumer-grade biometric devices have been reported to be as accurate as medical-grade products [
25,
26] and can be used for health care purposes [
27].
Two review papers [
19,
36] provide a list of consumer-grade EEG devices and a comprehensive literature survey for researchers’ reference. The key results are summarized below: four brands, namely, NeuroSky, Emotive, Interaxon, and OpenBCI, provide consumer-grade EEG products with high potential. Most EEG products are equipped with dry sensors, single channels, and BLE or classical Bluetooth communication protocols to stream data. The sampling frequency ranges from 128 to 512 Hz. The classification accuracy of the machine learning algorithm for the responded study ranges between 60% to 90%, as cited in the literature in the categories of cognition, education, entertainment, and brain–computer interaction. When comparing the performance of the power spectra, NeuroSky MindWave is similar to two medical-level EEG devices. Emotive EPOC is worse than the benchmark medical-level rival, Interaxon Muse, which demonstrates lower reliability than medical-level products. The results for OpenBCI resemble those of medical-grade devices. Moreover, a random time lag is a common phenomenon in wireless EEG devices.
Consumer-grade electroencephalography (EEG) products have been widely accepted for non-medical applications [
19,
36]. Similarly, PPG devices are popular for non-medical health care [
37,
38]. Innovations in technology over the last decade have advanced consumer-grade wearable sensors, enabling them to prosper in the market. The wireless, wearable, and lightweight features make consumer-grade products capable of collecting data continuously with no time or location limits. As a result, users do not feel discomfort after wearing such devices for a long period. In addition, they are easy to use and even novices can handle them easily.
In contrast, a medical-grade EEG system is accurate, but it comes at the price of a complex structure and long setup time [
36]. ECG does not offer users the flexibility, portability, or convenience offered by medical-grade EEG [
27]. Consequently, medical-grade device applications are usually limited to the laboratory [
19,
36].
Another point to note is that for research requiring high data retention, reliable communication, no real-time data transmission, and involving field investigations, wearable devices with temporary memory storage should be considered, with data uploaded to cloud servers at regular intervals. Examples of such research include investigations of the relationship between human well-being and daily experiences [
39] or the relationship between human emotions and daily life events [
40]. Physiological signals are excellent objective metrics for such research, and when combined with subjective assessments from participants, they can provide valuable insights. However, such studies involve scenarios that are not pre-designed or controlled; therefore, experiments can only be conducted in real-life settings. Carrying around another real-time data collection system is not feasible.
1.2. Relationship between User Experience and Users’ Emotional State
Norman coined and proposed the term “user experience” [
41]; however, initially, this term lacked a clear definition. Years later, Norman and Nielsen defined UX as “meeting the exact needs of the customer” and “a joy to own, a joy to use” [
42]. Some experts organized a special interest group (SIG) to comprehensively investigate UX. There are many definitions of UX in the survey results of the published literature or on the websites of UX organizations [
43]. The conclusion of the SIG suggests that researchers may choose their preferred definition from the identified list, which includes Norman and Nielsen’s definition.
The evaluation of UX traditionally uses participants’ self-evaluation through a questionnaire or personal interview after experiencing a product. This assessment method relies on the participants’ subjective perceptions, and is referred to as subjective evaluation. In contrast, UX evaluation based on biometrics is considered an objective method as it uses signals generated from the human autonomic nervous system. The SIG mentioned in the previous paragraph identified eighty-six UX evaluation tools [
44], but only four of these use physiological signals or facial expressions to assess UX. Therefore, thirty-three subjective methods utilize emotion/affect/hedonic as the index for UX evaluation. This finding is consistent with Norman and Nielsen’s definition, and implies that UX is related to a user’s emotional state after experiencing a product. Moreover, the limited number of tools using objective methods suggests that researchers should investigate this topic in greater depth.
1.3. Affective Computing as a Tool for User Experience Evaluation
Affective computing [
45] is an algorithm that recognizes a human’s emotional state [
46] through biometric signals [
47]. Accordingly, a user’s perceived pleasure while experiencing a product estimated using biometrics might serve as a UX metric. When a user experiences pleasurable emotions such as joy or positive valence for a product, it is reasonable to believe that this product generates a good user experience. Conversely, a user responding with anger or negative valence signals a poor user experience if. Instead of relying on a user self-report assessment or verbal expressions––referred to as subjective evaluation, and used in traditional UX studies––the affective computing applied in UX research is considered an objective method [
46]. Consequently, researchers interested in user experience (UX) use affective computing to recognize users’ emotional states when they experience a specific software or product [
48,
49,
50,
51,
52].
Although subjective methods are related to the assessment target, the participants may have cognitive bias [
53] and may not be sufficiently robust [
54]. An objective assessment could compensate for this disadvantage [
55]. Thus, subjective measures should not be the sole metrics used to evaluate UX [
49]. Using a subjective assessment of human emotion may be unreliable since emotions are often swift, hard to perceive, and sometimes have multiple states [
56]. Additionally, participants may be afraid to confess their emotions to the researcher. Worse, some participants may answer questions by imagining what the researcher expects them to say [
56].
In contrast, an objective metric could assist researchers to fill the gap caused by using subjective methods to evaluate the UX of a specific product. For example, a case study evaluated participants’ user experience of three different virtual dressing websites using verbal expressions and biometrics [
56]. There was no difference between the websites in terms of positive expressions resulting from verbal expressions. However, the percentage of engagement and attention, the positive/negative emotion, and the joy derived from biometrics showed that the three websites differed. This result illustrates the value of biometrics.
Moreover, affective computing can be applied in robotics to increase people’s enjoyment while interacting with robots. Physiological signals were one of the elements acting as medical robots’ human–machine interface. Using flexible electronics and devices makes the interface biocompatible, functional, conformable, and low-cost, resulting in an excellent user experience [
57]. Service robots in the health care sector can aid patients with cognitive obstruction via built-in affective computing algorithms [
58]. In commerce, empowering service robots with emotion recognition is a highly popular topic in research [
59,
60,
61,
62,
63,
64]. Some emotion recognition databases are available for service robots [
65,
66]. Accordingly, users’ emotional state during the experience with robots is a critical metric of human–robot interaction [
67,
68,
69]. Affective computing algorithms built into robots allow machines to recognize humans’ emotional states.
The present study focused on single-electrode electroencephalography (EEG), photoplethysmography (PPG) technology in heart rate (HR), galvanic skin response (GSR), and facial expressions, which are the biometrics frequently applied in affective computing studies [
4,
70,
71,
72,
73,
74,
75,
76,
77,
78,
79,
80,
81,
82,
83,
84].
Although subjective methods are commonly used for UX assessment, they have some limitations. For example, participants may have cognitive biases [
53] or the methods may not be sufficiently robust [
54]. To compensate for these limitations, objective assessments could be used [
55]. Therefore, subjective measures should not be the only metric used for UX evaluation [
49]. Assessing human emotions subjectively can also be unreliable since emotions can be swift, hard to perceive, and have multiple states [
56]. Additionally, participants may be reluctant to reveal their emotions to the researcher, and some participants may answer questions based on what they think the researcher expects them to say [
56].
1.4. Motivations and Objectives
This research aims to assist researchers who are interested in applying affective computing to investigate human behavior but are limited by a restricted budget for medical-grade instruments. While various consumer-grade alternatives are available, there are drawbacks to using these commercial products. Firstly, most consumer devices only provide a single biometric signal, which means that a study typically requires multiple physiological signals from different manufacturers. This means the software is not an integrated system, such as that used in medical-grade devices. Therefore, researchers must use multiple, specific software programs simultaneously to collect data. Consequently, each instrument may need an independent computer or mobile device, and software manipulation is more complex than with an integrated system. Moreover, it is impossible to synchronize data from different consumer alternatives in terms of timestamps, even with multiple experimenters collaborating to press the buttons simultaneously. This can lead to a minimal time gap of less than one second, which may be insufficient for some psychology studies. Finally, the data processing of different signals distributed in many files can also be troublesome.
The objective of this research is to use affordable hardware and free software libraries to develop a low-cost biofeedback platform for biometrics-related studies, such as human factors engineering, user experience, human behavioral studies, and human–robot interaction. The specific aim of the hardware is achieved by using consumer products or electronic modules designed for Arduino makers rather than medical-grade instruments. Another objective of the software is achieved through the use of open-source libraries that can be used free of charge. The software is an object-oriented programming (OOP) class that serves as a software development kit (SDK), which can be reused for other research or as a standalone biometrics collecting system. An SDK design allows researchers to integrate data collection with their experimental stimulus into a single system through customized coding. Nonintegrated devices require at least one additional computer beside the one controlling the experiment stimulus [
85]. Moreover, the integrated software simplifies researchers’ manipulation during experimentation. A single button click triggers the stimulus, and the data collection is synchronized. Therefore, the biofeedback platform developed in this study improves the ease of operation in experiments and reduces equipment costs.