Robotic Delivery Worker in the Dark: Assessment of Perceived Safety from Sidewalk Autonomous Delivery Robots’ Lighting Colors

Abstract

Sidewalk autonomous delivery robots (SADRs) share sidewalks with pedestrians and may affect their perceived safety. In outdoor nighttime environments, the color of the lights on SADRs serves as a noticeable form of communication that can influence human emotions. Therefore, this study investigated whether the perceived safety of SADRs varied with the colors of their lighting. In this study, an experiment (n = 30) was conducted where participants walked towards a robot from opposite directions to interact. The perceived safety of five different lighting colors (red, yellow, blue, green, and white) on SADRs was measured before and after the interaction using two perceived safety scales. The results showed significant differences in participants’ perceived safety for the robot’s red, green, blue, and yellow lighting before and after the interaction. Red lighting was rated the least perceived safe, while white and yellow lighting were rated the most perceived safe. Additionally, gender significantly influenced the perceived safety assessments, with females reporting lower perceived safety than males. These findings are valuable when designing SADRs that enhance pedestrians’ perceived safety, thereby facilitating their integration into broader environments in the future.

1. Introduction

In recent years, the application domains of service robots have continuously expanded, encompassing manufacturing, agriculture, healthcare, logistics, retail, banking, hospitality, smart cities, and public sectors [1]. These robots have played a crucial role in enhancing efficiency, particularly the sidewalk autonomous delivery robots (SADRs) that are employed in last-mile delivery, which are considered to be the latest development in service robotics [2,3]. SADRs are small robots that autonomously navigate sidewalks without human intervention to deliver food or goods and can be called a “robotic delivery worker” [4]. Unlike service robots used indoors in malls and hotels, SADRs operate on sidewalks, sharing the space with pedestrians [5]. Previous research has found that SADRs may impact pedestrians’ physical safety [6,7], prompting attention on navigation planning to enhance pedestrians’ physical safety to prevent physical harm [8]. However, scant research has considered pedestrians’ perceived safety [9,10], which refers to not causing harm to pedestrians’ psychological wellbeing, posing additional challenges for the human–robot interaction (HRI) of SADRs [11].

The field of HRI examines factors that influence perceived safety, including robot factors, environmental factors, and human factors. Robot factors encompass the robot’s appearance, lighting, sound, movements, and other characteristics. Among these, lighting is a significant research area; it can be emitted by LEDs or other light sources, and different lighting colors can subtly influence human perception and psychology, even affecting human behavior [12]. For robots that cannot make physical gestures and lack display screens [13], the lighting color serves as an essential means of communication. Given that SADRs undertake delivery tasks at night, with increased opportunities for interactions between pedestrians and SADRs following wider commercialization, the further exploration of SADRs’ lighting colors is crucial.

Environmental and human factors also merit consideration in this study. Due to the nature of SADRs’ delivery tasks, both pedestrians and SADRs are mobile during interactions. In outdoor environments and dark conditions, an individual’s perceived safety may differ from that in confined indoor spaces. Therefore, it is necessary to investigate interactions between pedestrians and SADRs in walking scenarios and under dark outdoor conditions. Additionally, Nomura et al. [14] confirmed that male participants generally exhibit a more positive attitude towards robot interactions than female participants, which this study also aims to explore.

This study aims to evaluate pedestrians’ perceived safety regarding SADRs with different lighting colors in dark outdoor environments. Specifically, this study seeks to answer three research questions: (1) Is a perceived safety evaluation related to pedestrians’ interactive behavior such as walking? (2) Are there differences in perceived safety among pedestrians for different lighting colors? (3) Does gender influence perceived safety assessments?

2. Literature Review

2.1. Sidewalk Autonomous Delivery Robots

SADRs are typically defined as robots that autonomously navigate sidewalks for a certain distance and adhere to traffic regulations to deliver food or items to consumers [15,16]. The delivery range of SADRs primarily covers outdoor environments such as parks, campuses, and residential apartments. Equipped with sensors and navigation technology, they generally travel at a walking pace [4]. Their payload capacity typically ranges from 10 to 30 kg, allowing for single-trip deliveries only, thus preventing the possibility of simultaneously serving multiple customers. SADRs belong to the category of non-humanoid robots [17], meaning that they lack a human-like appearance or structures such as limbs and facial features, and are commonly used for specific tasks. To ensure stability during delivery and to accommodate a certain capacity, SADRs are typically designed to resemble a mobile box with wheels.

SADRs have also gained attention in South Korea. According to data from the first half of 2022, SADRs accounted for 2.8% of the search volume on Naver and Google, highlighting their prominence in the logistics field [18]. Emerging companies such as Neubility, Woowa Brothers, CJ Logistics, and ROBOTICS have designed and manufactured SADRs. Neubie, Dilly Drive, Gaemi, and Mobin are four SADRs currently operating or under trial in South Korea, as shown in Figure 1. The South Korean government has taken measures to support the commercialization of SADRs, formally implementing the “Amendment to the Act on the Promotion of Development and Distribution of Intelligent Robots” in November 2023, allowing SADRs to enter public spaces. Given this societal background, there is an urgent need for research on HRI in SADRs.

2.2. Perceived Safety in HRI

Safety is one of the basic human needs [19]. In the realm of cognitive human–robot interaction (cHRI), perceived safety is a crucial concept. It refers to the cognitive state wherein individuals do not experience discomfort or stress due to the visual and behavioral characteristics of the robot during HRI; rather, they feel confident and secure [20]. Perceived safety is often described in terms of positive emotional states such as trust, ease [21], and comfort [22] or negative emotional states such as stress [23], danger [24], fear [25], and anxiety [26].

Although considerable research has focused on physical safety in HRI to prevent harm to humans, there has been a tendency to overlook perceived safety. Simply preventing collisions to maintain physical safety may still result in a lower level of perceived safety [27]. Perceived safety is essential for long-term interaction, collaboration, and acceptance between humans and robots, making it indispensable to ensure individuals’ perceived safety during interactions with robots [28].

Existing research on SADRs has revealed the potential to generate safety concerns among individuals [5,29], highlighting the need to identify the factors influencing the perceived safety of SADRs. However, modeling the factors is a challenging task. Attributes of the robot such as appearance (e.g., size, shape, posture, etc.) and movements (e.g., speed, acceleration, proximity to humans, etc.) influence perceived safety. Given the bidirectional nature of HRI involving both humans and robots, perceived safety cannot be solely based on robot attributes [30]. Human factors as well as physical environmental factors in which humans and robots coexist must also be considered.

2.3. Robot Lighting Color and Human Emotions

Lighting is suitable for long-distance transmission; it is easily recognizable and capable of withstanding higher levels of visual degradation. The human color-vision system processes color signals in a distinct manner [31], eliciting emotional responses to colors. Consequently, in the field of HRI, many robots use expressive colored lighting as visual cues for interactions [32]. Colored lighting serves as a form of non-verbal communication through the visual channel and is capable of conveying states, expressing emotions, or enhancing attractiveness [33]. When robots employ colored lighting, these combinations of colors significantly impact on users’ emotions, cognition, and attention [34].

Colors typically include warm tones, cool tones, and neutral tones [35]. Warm tones generally encompass red and yellow, cool tones include blue and green, and neutral tones include black and white. Humans have different perceptions of these three lighting colors when used on robots. Muthugala et al. [36] found that the lighting color of floor-cleaning robots could signal attention to users, making robots with lights appear more user-friendly. When the light color was between yellow and red, the floor-cleaning robot elicited a moderate level of attention from humans. Dou et al. [37] mentioned that when medical robots used warm-colored lighting (yellow) as feedback, participants perceived it to be warmer compared with neutral lighting (white) and cool-colored lighting (blue), which caused strong discomfort among participants. Song and Yamada [38] found that people prefer blue, green, and yellow as feedback from robots. Blue lighting makes people feel comfortable and safe [39]. Dou et al. [40] found that among several lighting colors of social robots, neutral tones were the most accepted. However, SADRs often operate at night and outdoors, and how their lighting colors affect human emotions—thereby influencing perceived safety—remains an issue that current research has yet to address.

3. Materials and Methods

To obtain participants’ perceived safety assessments of SADRs, this study used an experimental robot and five lighting colors: white, green, red, blue, and yellow. A within-subject design was employed wherein participants were exposed to all experimental variables. This approach was chosen to allow for a direct comparison of participants’ perceived safety assessments using different experimental stimuli for the same individual. To accurately capture participants’ real feelings, the experiment was conducted in outdoor environments at night, with participants completing a survey questionnaire. Subsequently, the data obtained from the questionnaire were statistically analyzed.

3.1. Experimental Robot and Lighting-Color Representation

To achieve mobility of the experimental robot and variations in lighting colors, an LIMO robot (Agilex robotics) was used as the base. This multi-mode, compact, and customizable autonomous mobile robot was chosen for its versatility. Python programming was employed to control the robot’s speed and travel route for navigation planning. To resemble the appearance and size of the Neubie robot [41] (dimensions: width, 53 cm; length, 61 cm; and height, 69 cm), a plastic shell was installed on the exterior of the LIMO robot to achieve a 1:1 effect. LED light strips with a luminous efficacy of 220 lm/w were attached to the shell of the experimental robot, with the lighting colors controlled using a remote controller. The appearance of both the LIMO robot and the experimental robot are shown in Figure 2.

3.2. Experimental Variables

Based on the literature review, red and yellow are categorized as warm colors, while blue and green are considered to be cool colors and white serves as a neutral color. Thus, the variable of the lighting color, which served as the experimental stimulus, was controlled to include five colors: red, yellow, blue, green, and white. The illumination mode was set to constant light. The lighting-color representation of the experimental robot is shown in

Table 1. Five lighting colors of the experimental robot.

White	Green	Red	Blue	Yellow

.

To minimize the influence of irrelevant variables on perceived safety, both the robot and the participants followed a fixed route for movement. The experimental site was divided into a test zone (5 m × 2 m) and a control zone. Both started from the ends of the test zone at a starting-point distance of 5 m. According to Hall [42], personal space is defined as 45 cm–120 cm, which is the distance at which humans converse with friends. Hüttenrauch et al. [43] confirmed that people allow robots to enter their personal space. Therefore, the lateral distance between the robot and the participants was controlled at 90 cm. The robot traveled at a slow speed of 40 cm/s, which is the speed Neubie robots use when encountering pedestrians. The researchers and experimenters stood in the control area, which was 2 m from the test area, to control the robot and guide the participants. A schematic diagram of the experiment is shown in Figure 3.

3.3. Perceived Safety Scales

Various scales have been developed in the field of HRI to measure perceived safety, including the Godspeed Questionnaire (GSQ) [44], NARS [45], and RoSAS [46] as well as scales designed by Akalin et al. [47]. The Akalin et al. [47] perceived safety scale is a five-point Likert semantic differential scale consisting of eight pairs of opposite semantic terms: insecure–secure, anxious–relaxed, uncomfortable–comfortable, lack-of-control–in control, threatening–safe, unfamiliar–familiar, unreliable–reliable, and scary–calming. The threatening–safe, unfamiliar–familiar, unreliable–reliable, and scary–calming pairs of attributes are used to measure people’s initial perceptions of the robot. The threatening–safe, anxious–relaxed, uncomfortable–comfortable, and lack-of-control–in control pairs are used to assess people’s feelings during interactions with the robot.

To compare the impact of the interaction behavior on perceived safety, this study referenced the Akalin et al. [47] perceived safety scale and designed two types of scales: a pre-interaction perceived safety scale and a post-interaction perceived safety scale. The pre-interaction perceived safety scale included four pairs of descriptive terms: threatening–safe, unreliable–reliable, scary–calming, and unfamiliar–familiar. The details of this scale can be found in

Table 2. Pre-interaction perceived safety scale.

No.	Negative Connotation	Positive Connotation
1	Threatening	Safe
2	Unreliable	Reliable
3	Scary	Calming
4	Unfamiliar	Familiar

.

The post-interaction perceived safety scale included seven pairs of descriptive terms: anxious–relaxed, uncomfortable–comfortable, lack-of-control–in control, threatening–safe, unfamiliar–familiar, unreliable–reliable, and scary–calming. The details of this scale can be found in

Table 3. Post-interaction perceived safety scale.

No.	Negative Connotation	Positive Connotation
1	Threatening	Safe
2	Unreliable	Reliable
3	Scary	Calming
4	Unfamiliar	Familiar
5	Anxious	Relaxed
6	Uncomfortable	Comfortable
7	Lack of control	In control

. The insecure–secure pair was not used in the measurements after discussions with two interaction design experts as its meaning is similar to threatening–safe, which could cause confusion. Each scale utilized a five-point Likert measurement, where 1 indicated threatening, 3 indicated neutral, and 5 indicated safe.

3.4. Participant Recruitment

In total, 30 students from a university campus were recruited as participants for this study, including 16 males and 14 females aged 22 to 39 years (M = 27; SD = 3.303). The sample comprised 27 Chinese and 3 Korean participants. Each participant completed a pre-test questionnaire that collected data on age, gender, educational background, health status, and if they had any color blindness or color-vision deficiency. The results confirmed that all participants met the requirements for the experiment. Notably, 17 of the participants were from an industrial design program, giving them familiarity with robot design principles and a higher acceptance of SADRs.

The implementation procedures of the experiment were approved by the Hanyang University Ethics Committee and complied with the 1964 Declaration of Helsinki and its subsequent amendments. Before the start of the experiment, participants provided written informed consent. They were thoroughly debriefed after each session.

3.5. Experimental Procedure

The experiment was conducted from 2 December to 6 December 2023, during the evenings (6:30 pm–9 pm). The venue was an open space within the college of design at Hanyang University. The area was spacious, devoid of external disturbances, and lit by streetlights on both sides. The ground comprised a flat wooden floor, closely resembling the actual usage environment of SADRs. The specific steps of the experiment were as follows:

• Participants voluntarily signed an informed consent form in the office, where the researcher recorded their identities.
• Participants entered the experimental area, and the researcher explained the experimental procedure to them.
• The researcher guided the participants to the starting point to observe static robots with different lighting colors for 10 s each. After each observation, participants completed the pre-interaction perceived safety questionnaire (Table 2), repeating this process five times.
• Another experimenter set the robot’s speed and route, ensuring the robot and participant moved simultaneously.
• The researcher guided the participants to walk along a fixed route in the opposite direction to moving robots with different lighting colors, repeating this process five times.
• After each walk, participants were required to fill out the post-interaction perceived safety questionnaire (Table 3), repeating this process five times.
• Upon completing the outdoor experiment, the experimenter collected all questionnaire responses and invited participants back to the office for a 5 min interview with another experimenter.

After completing all the experiments, each participant received a small gift as a reward. Throughout the entire experimental process and data handling, participants’ privacy was protected.

4. Results

4.1. Differences in Participants’ Perceived Safety of Robot Lighting Colors Pre-Interaction and Post-Interaction

The differences in perceived safety between pre-interactions and post-interactions were evaluated based on four questionnaire items: threatening–safe, unreliable–reliable, scary–calming, and unfamiliar–familiar. Descriptive statistics and a paired sample t-test were used to examine the differences in these four perceived safety assessments before and after interactions, as shown in Figure 4 and

Table 4. Paired sample t-test results.

Paired Samples		Paired Samples (Mean ± SD)		Mean Difference	t	p
		Pre-Interaction	Post-Interaction
White lighting	Safe	4.133 ± 0.86	4.033 ± 0.999	0.1 ± −0.139	1	0.326
	Reliable	4.2 ± 0.714	4.167 ± 0.699	0.033 ± 0.015	1	0.326
	Calming	4.1 ± 0.885	3.933 ± 0.944	0.167 ± −0.06	1.98	0.057
	Familiar	3.467 ± 0.681	3.5 ± 0.682	−0.033 ± −0.001	−0.441	0.662
Red lighting	Safe	1.767 ± 0.679	1.633 ± 0.718	0.133 ± −0.039	1.278	0.211
	Reliable	2 ± 0.695	1.9 ± 0.759	0.1 ± −0.064	1.795	0.083
	Calming	1.833 ± 0.648	1.667 ± 0.661	0.167 ± −0.013	2.408	0.023 **
	Familiar	2.6 ± 0.621	2.533 ± 0.819	0.067 ± −0.198	0.701	0.489
Green lighting	Safe	2.567 ± 0.935	2.167 ± 0.913	0.4 ± 0.022	4.397	0.000 ***
	Reliable	2.6 ± 0.932	2.4 ± 0.932	0.2 ± 0	2.693	0.012 **
	Calming	2.567 ± 0.971	2.267 ± 0.907	0.3 ± 0.064	3.071	0.005 ***
	Familiar	2.8 ± 0.847	2.667 ± 0.959	0.133 ± −0.112	1.682	0.103
Blue lighting	Safe	3.6 ± 0.894	3.167 ± 0.95	0.433 ± −0.055	2.904	0.007 ***
	Reliable	3.533 ± 0.937	3.3 ± 1.055	0.233 ± −0.118	1.564	0.129
	Calming	3.467 ± 0.937	3 ± 0.871	0.467 ± 0.066	3.294	0.003 ***
	Familiar	3.2 ± 0.551	3.233 ± 0.568	−0.033 ± −0.017	−0.372	0.712
Yellow lighting	Safe	4.5 ± 0.63	4.333 ± 0.711	0.167 ± −0.081	1.98	0.057
	Reliable	4.567 ± 0.568	4.433 ± 0.679	0.133 ± −0.111	1.682	0.103
	Calming	4.6 ± 0.563	4.267 ± 0.691	0.333 ± −0.128	3.01	0.005 ***
	Familiar	3.7 ± 0.651	3.7 ± 0.702	0 ± −0.051	0	1

Note: ** p < 0.01, and *** p < 0.001.

.

The descriptive statistics revealed that across all dimensions, the evaluations ranked from low to high as follows: red, green, blue, yellow, white. The evaluations for yellow and white were similar. Specifically, in the dimensions of threatening–safe, unreliable–reliable, and scary–calming, the five lighting colors exhibited similar perceptions. However, in the dimension of unfamiliar–familiar, red and green remained moderate, while white, yellow, and blue maintained higher levels of familiarity.

The paired sample t-test results revealed no significant differences in perceived safety across all dimensions for white lighting before and after interactions. However, for the scary–calming dimension, participants felt significantly more scared after interactions when the lighting color was red (M = 0.167, SD = −0.013, and p < 0.05), green (M = 0.3, SD = 0.064, and p < 0.05), blue (M = 0.467, SD = 0.066, and p < 0.05), or yellow (M = 0.333, SD = 0.128, and p < 0.05). For the threatening–safe dimension, participants felt significantly more threatened after interactions when the lighting color was green (M = 0.4, SD = −0.022, and p < 0.05) or blue (M = 0.433, SD = 0.055, and p < 0.05). Additionally, participants felt significantly more unreliable after interactions when the lighting color was green (M = 0.2, SD = 0, and p < 0.05).

4.2. Differences in Participants’ Perceived Safety of Different Robot Light Colors Post-Interaction

A one-way analysis of variance (one-way ANOVA) was used to examine the discrepancies in the participants’ perceived safety of five different robot light colors after interactions, as illustrated in

Table 5. One-way ANOVA results.

Lighting Color (Mean ± SD)					F	p
White Lighting	Red Lighting	Green Lighting	Blue Lighting	Yellow Lighting
3.908 ± 0.677	1.933 ± 0.619	2.375 ± 0.814	3.175 ± 0.717	4.183 ± 0.517	61.007	0.000 ***

Note: *** p < 0.001.

. The results indicated notable differences in perceived safety among the five light colors (F = 62.325; p < 0.05), suggesting that post hoc analyses could be conducted.

Tukey post hoc tests were executed to compare participants’ perceived safety regarding the five different robot lighting colors, as depicted in

Table 6. Tukey post hoc test results.

Lighting Color 1	Lighting Color 2	Lighting Color 1 Mean	Lighting Color 2 Mean	Mean Difference	p
White	Red	3.908	1.933	1.975	0.000 ***
White	Green	3.908	2.375	1.533	0.000 ***
White	Blue	3.908	3.175	0.733	0.000 ***
White	Yellow	3.908	4.183	−0.275	0.117
Red	Green	1.933	2.375	−0.442	0.012 **
Red	Blue	1.933	3.175	−1.242	0.000 ***
Red	Yellow	1.933	4.183	−2.25	0.000 ***
Green	Blue	2.375	3.175	−0.8	0.000 ***
Green	Yellow	2.375	4.183	−1.808	0.000 ***
Blue	Yellow	3.175	4.183	−1.008	0.000 ***

Note: ** p < 0.01, and *** p < 0.001.

. The findings indicated significant variations in perceived safety between white (M = 3.908) > red (M = 1.933), white (M = 3.908) > green (M = 2.375), white (M = 3.908) > blue (M = 3.175), blue (M = 3.175) > red (M = 1.933), yellow (M = 4.183) > red (M = 1.933), green (M = 2.375) > red (M = 1.933), blue (M = 3.175) > green (M = 2.375), yellow (M = 4.183) > green (M = 2.375), and yellow (M = 4.183) > blue (M = 3.175), all showing significance (p < 0.05). However, no significant difference was observed between white and yellow (p > 0.05). This indicated that red (M = 1.933) elicited the lowest perceived safety, while white (M = 3.908) and yellow (M = 4.183) elicited the highest perceived safety.

4.3. Differences in Perceived Safety of Robot Lighting Colors among Participants of Different Genders

Independent sample t-tests were employed to scrutinize variances in the perceived safety of robot lighting colors between male and female participants after interactions, as shown in

Table 7. Sample t-test results.

Lighting Color	Participant Group (Mean ± SD)		t	p
	Male (n = 16)	Female (n = 14)
White	3.969 ± 0.591	3.839 ± 0.782	0.516	0.618
Red	2.203 ± 0.579	1.625 ± 0.526	2.847	0.008 **
Green	2.766 ± 0.803	1.929 ± 0.575	2.236	0.003 **
Blue	3.5 ± 0.387	2.804 ± 0.833	2.999	0.006 **
Yellow	4.109 ± 0.577	4.268 ± 0.444	−0.834	0.411

Note: ** p < 0.01.

. The results indicated noteworthy differences in perceived safety between males and females for red, green, and blue lights (t > 2; p < 0.05), with females attributing significantly lower perceived safety to these three types of lights compared with males. Nevertheless, there was no significant difference in perceived safety between males and females for white and yellow lights (p > 0.05).

4.4. Interview Results

The interview transcripts of participants were analyzed to capture their perceptions of the five lighting colors.

4.4.1. Negative Associations with Red Lighting

Participants frequently associated red light with danger, akin to red lights in traffic signals. In total, 70% of the participants expressed aversion and fear towards the red light. Participant C05 remarked, “The robot with red lights approached me, and I became wary”. Participant C03 stated, “I find red light unacceptable; it feels like car tail lights signaling danger, urging me to leave”.

4.4.2. Mixed Reactions to Green and Blue Lighting

Regarding the green light, participants experienced both negative and positive emotions. In total, 53% of the participants found the green light strange, expressing feelings of fear and anxiety. Participant C08 commented, “Green light looks odd; if a green light robot passes by me outdoors at night, I would be scared”. Participant C22 said, “Green light seems to signal permission, like the green light in traffic signals. Seeing green light makes me feel safe walking alongside the robot”.

Similarly, participants held diverse opinions about the blue light. Although 46% of the participants found the blue light intimidating, 20% of the participants associated it positively with technology and intelligence. Participant C25 mentioned, “Blue light is somewhat intimidating; cold-toned lights make me uneasy”. In contrast, Participant C16 noted, “Blue light represents technology and intelligence; I don’t feel uneasy about it”.

4.4.3. Positive Views on White Lighting

In total, 73% of the participants unanimously regarded the white light as the most soothing. Participant C01 commented, “White light is the most comfortable for me because it doesn’t carry any specific meaning; it just shows where the robot is”. Participant C15 echoed this sentiment, “White light is good; it helps me see the robot clearly and the ground”.

4.4.4. Warm Reception towards Yellow Lighting

In total, 76% of the participants expressed a preference for the yellow light, describing it as warm and comfortable. However, some participants associated the yellow light with the cautionary meaning of a yellow traffic signal. Participant C30 said, “Yellow light looks warm, like streetlights”. Participant C21 observed, “Isn’t yellow light supposed to mean ‘wait’, like the yellow light in traffic signals?”. Participant C04 remarked, “Watering trucks have yellow lights because they move slowly. If robots use yellow lights, it should mean they are moving slowly”.

5. Discussion

SADRs will become road-sharing companions for pedestrians at night, making perceived safety a critical factor. Understanding the factors that could diminish perceived safety, such as light colors, is essential when designing suitable robots. This study investigated the impact of five light colors on participants’ perceived safety.

5.1. Impact of Interaction Behavior on Perceived Safety

Participants’ perceived safety concerning the color of robot lights differed before and after interacting with the robots, particularly for the colors red, green, blue, and yellow. During the experiment, participants experienced realistic interactions as the robots moved closer and passed by them. Takayama and Pantofaru [48] suggested that perceived safety is inversely related to the robot’s proximity, a finding supported by this study. Additionally, Arai et al. [49] noted that higher robot speeds induced greater stress compared with lower speeds. Although this study did not include both fast and slow robot speeds, it revealed that even at a slow speed, robots were perceived to be less safe when in motion compared with when they were stationary.

5.2. Relationship between Lighting Colors and Perceived Safety

Participants exhibited varying levels of perceived safety for robots with different light colors. In color psychology, red can evoke extreme emotions, both excitement (positive) and fear (negative). Song and Yamada [13] discovered that red could signify robot hostility. Consistent with previous findings, this study revealed that a red light elicited the lowest perceived safety, with participants displaying significant aversion and fear towards it. Although blue is generally considered to be pleasant, and a low-intensity blue light on robots was found to be highly attractive in the research of Song and Yamada [13], this study did not reflect the same phenomenon. A blue light led to lower perceived safety, aligning with the findings of Dou et al. [37] that cool-colored lights cause discomfort. In outdoor environments, a blue light is rare and is less acceptable when seen on moving robots, a phenomenon that also applies to green light.

White light is considered to be suitable visual feedback for robots in any context [37], a view supported by this study. Participants consistently rated a white light with high perceived safety, regardless of whether the robot was moving or stationary. Dou et al. [40] suggested that a yellow light (warm-colored light) on social robots did not easily cause discomfort, a finding corroborated by this study.

Robots need to attract human users to establish successful interactions while deterring unfriendly individuals to avoid abuse [13]. This insight, combined with the varying attitudes towards different light colors, suggests design strategies for SADRs. For instance, using red light to signal pedestrians to maintain a distance could be an effective approach.

5.3. Relationship between Human Factors, Environmental Factors, and Perceived Safety

Gender is an individual characteristic that can influence perceived safety [27]. For research question 3, this study investigated the role of gender in perceived safety. The findings of this study indicated that male participants exhibited higher perceived safety and greater comfort with red, green, and blue lighting colors compared with female participants. Both yellow and white lighting, regardless of gender, showed higher perceived safety. The reasons for this difference may have been a combination of male and female attitudes towards robots or their varying sensitivities to colors. This would align with previous studies, showing that gender affects attitudes and anxiety towards robots [14]. Male participants are more proactive and feel safer during interactions [30]. Women can recognize a broader range of colors than men [50] and are more discerning when distinguishing between red and green hues [51]. The results of this study broaden these perspectives by highlighting the differences in perceived safety between men and women for red, green, and blue light colors.

Additionally, the environment plays a crucial role in people’s reactions to specific stimuli. In this context, the environment includes both the physical surroundings and the type of task the robot is performing. Participants frequently associated the robot’s red, green, and yellow lights with traffic signals or car tail-lights. This association may have been related to the experimental environment and the robot’s delivery task, both of which were reminiscent of road traffic conditions.

5.4. Limitations and Future Work

Despite the contributions of this study, several limitations should be acknowledged. Firstly, the study recruited 30 young student participants who may have been generally more familiar with and accepting of robots compared with older populations. Consequently, they exhibited higher perceived safety towards the experimental robots. However, in real road environments, vulnerable groups such as the elderly, children, and individuals with disabilities are also important pedestrians. Their walking speeds, heights, and cognitive abilities differ from those of healthy young adults, and they may have different opinions about SADRs. Future experiments should expand the range of participants to include comparisons between young and elderly groups or focus on the opinions of vulnerable groups regarding SADRs.

Secondly, this study used LED light strips and custom-made enclosures to simulate the experimental robots, with the robots’ movements entirely controlled by the researchers. This setup did not fully replicate the behavior of real SADRs in uncontrolled environments. As Nyholm et al. [52] noted, perceived safety is influenced by the presence of human assistants during HRI. Participants may perceive controlled, non-realistic robots as safer. Thus, future research should utilize real SADRs and explore more experimental parameters while minimizing the researchers’ presence to obtain more comprehensive results.

Thirdly, in this study, the robot generated noise from its wheels whilst moving, which became more pronounced as it approached the participants. This suggests that the sound signaling the robot’s approach to the participants may have also been a factor influencing perceived safety. In future research, the impact of robot noise on perceived safety is an important topic worth exploring.

Fourthly, the cultural background can influence color perception. The participants in this study included both Koreans and Chinese, who may have different color preferences. For instance, white is more favored by Koreans, whereas Chinese participants may prefer red [34]. Future research could benefit from conducting cross-cultural comparisons to extend the findings of this study.

6. Conclusions

This study investigated the lighting colors of SADRs and examined human perceived safety assessments of these colors. The findings revealed that participants’ perceived safety decreased after interacting with SADRs when the lighting colors were red, green, blue, or yellow. This suggests that dynamic interactions such as walking alongside SADRs negatively impact perceived safety, unlike static experiences. After interactions, red lighting resulted in the lowest perceived safety. Green lighting was also poorly rated, while yellow and white lighting were rated the highest for perceived safety. Therefore, SADRs should avoid using red and green lighting, with white and yellow being the optimal choices. Additionally, there were significant differences in perceived safety between genders, with women reporting lower perceived safety than men for the same lighting colors. Consequently, SADRs should be designed with light colors that are acceptable to both men and women.

The findings from this study enhance our understanding of humans’ perceived safety in relation to SADR lighting colors. By adjusting the lighting colors of SADRs, their interactive behavior can be optimized to improve human acceptance of the robots. We hope the results of this study can serve as a reference for other similar perceived safety assessments. As discussed, further research is needed to continue exploring these findings.