Embo: A Wearable Robot Transforming Child-Directed Verbal Aggression into Tactile Feedback

Verbal aggression in child development is a pervasive issue, impacting both victims and aggressors profoundly. Traditional approaches, such as cognitive education and reactive strategies, lack practicality and intuitive understanding, presenting limitations in addressing this problem effectively. In response, we propose an innovative solution: "Embo," a wearable hand puppet robot that translates verbal aggression intensity into a tactile discomfort, offering a preventative tool for children. This novel intervention aims to mitigate the harmful effects of verbal aggression, contributing to a safer and healthier environment for children.


INTRODUCTION
Verbal aggression, defned as the act of inficting humiliation, denigration, and fear through spoken words, is a signifcant issue in child development.Children often encounter such aggression from peers, teachers, and parents, and they gradually learn to adopt verbal aggression themselves.Studies show that 71% of teachers observe daily verbal aggression among students [16], and most parents also engage in such behavior [13].This aggression results in lasting psychological harm: victims sufer from depression, anxiety, academic challenges, and low self-esteem [2,13], while aggressors may face social exclusion and delinquency [10].
Traditional psychological interventions focus on preemptive cognitive education and reactive strategies during incidents of aggression.These interventions typically involve theoretical instructions and role-playing to enhance awareness of the negative impacts of child-directed verbal aggression [9,19].However, they often lack practical application and fail to ofer an intuitive understanding of the harms caused by verbal aggression.Reactive strategies, such as engaging in alternative activities to divert attention [20] or using metaphorical scales to monitor emotional escalation [7], overly rely on self-awareness and control, showing limited efectiveness for those lacking self-insight.Researchers in the feld of robotics also have attempted to address conficts and aggressive encounters using peripheral interventions.Notable examples include robots programmed to shiver as an indication of fear in response to aggressive conversations [4] or robots that use voice commands to intervene or suggest solutions upon detecting conficts [6,12].While these peripheral interventions address the over-reliance on self-awareness in psychological strategies, they may not efectively de-escalate conficts in high-stakes or emotionally charged situations due to the limited impact of peripheral cues or may intervene too abruptly, failing to motivate users to change their behavior.

Figure 1: A schematic of the wearable interactive robot
Therefore, we introduce Embo (Figure 1), a hand puppet robot designed to translate the intensity of child-directed verbal aggression into tactile feedback.Embo acts as the recipient of verbal aggression.Upon receiving aggressive expressions, it evaluates their severity and adjusts its airbag infation accordingly.The diferent levels of airbag pressure, as it squeezes the users' hands, enable users to feel the emotional distress that Embo experiences.This innovative approach ofers tangible representations of Embo's emotional states, thereby aiding users in understanding the negative impacts of verbal aggression and developing empathy towards its victims in real life.

DESIGN CONSIDERATIONS 2.1 Conceptual Foundation
Our robot is innovatively designed to recognize aggressive speech through voice recognition technology and provides tactile feedback to alert users about the occurrence of verbal aggression behavior.This feature is instrumental in enhancing self-awareness among users.Research indicates that individuals with heightened selfcontrol and refective capacities exhibit fewer instances of verbal aggression [15].Therefore, utilizing our robot for self-awareness training in recognizing child-directed verbal aggression can efectively ameliorate issues related to language-based aggression.
Additionally, the robot's anthropomorphic design assists aggressors in understanding and empathizing with the victim's level of distress.Humans possess an innate ability to anthropomorphize, projecting emotions onto inanimate objects [1].Thus, users can naturally correlate the robot's emotional expressions with the children subjected to verbal aggression.It is noteworthy that children, at ages earlier than 4 years, have already developed the ability to perceive and understand others' internal states, including negative emotions [14].Moreover, preschool children can perceive robots as sentient and emotional beings and show a positive willingness and behavior for interaction [17].Therefore, under adults' guidance, children can efectively engage with our device, achieving our intention of mitigating verbal aggression.

Considerations for Industrial Design
Embo's industrial design thoughtfully incorporates human, machine, and environmental elements, presenting a hand-worn robot that enhances user interaction and provides intuitive feedback.Embo is crafted using elastic fabrics and standardized sizing to ensure adaptability, while plush materials are employed to foster trust and comfort [18].The robot's anthropomorphic design, resembling human form and movements, strengthens emotional connections and empathy for victims of verbal aggression.Embo ofers a unique, engaging experience that boosts user interest and acceptance, enhancing its efectiveness in training against verbal aggression.

Grading of Verbal Aggression
In the context of verbal aggression, the varying impact of diferent expressions on the language target necessitates hierarchical diferentiation of verbal aggression.This enables the robot to provide tactile feedback at diferent levels.
The categorization of verbal aggression is closely tied to its level of harm, and analyzing the aggressiveness of language requires Table 1: Grading and cases of verbal aggression attention to the specifc content conveyed by the aggressor.Hughes conducted a study on English abusive language, categorizing it based on the social application of vocabulary, including blasphemy, obscenity, and categories targeting women and race [5].Minjie Li associated the aggressiveness of language in the context of online abuse with the psychological state of the language target, providing a novel perspective [8].
Considering limited resources to ensure a better user experience, we ultimately defne the interaction hierarchy as 3, incorporating previous approaches to verbal aggression categorization.Firstly, abusive language is preliminarily divided into three levels: (1) Level 1, expressing strong derogatory remarks about others, such as "you're really ugly." (2) Level 2, expressing strong feelings of disgust and escalating to personal insults, like "you're useless." (3) Level 3, directly attacking and insulting the personality of the other, such as "why don't you go die." Subsequently, we introduce the language target's three-tier psychological response to verbal aggression, forming the fnal verbal aggression categorization as illustrated in Table 1.Our semantic library is extensive, encompassing common phrases related to child-directed verbal aggression.
In order to refne the meaning of the aids, we also included the apology vocabulary in the semantic base, but the soothing efect of apologies on psychological trauma is a new and large topic, so we will not explore it in depth.Our robot features the appearance of a hand puppet, with key components including an integrated microphone, an airbag, as well as an external control unit and two air pumps.This thoughtful design empowers the robot with the ability to dynamically respond to varying degrees of child -directed verbal aggression.The detailed technical breakdown is illustrated in Figure 2.

Input
When the wearer interacts with the puppet robot, the robot's builtin microphone captures the wearer's speech.Furthermore, we utilize state-of-the-art Speech-to-Text technology (Whisper API) to transcribe the speech into textual information, serving as input for the control system.

Control System
We utilize GPT-4, a large language model (LLM), to process these textual data for in-depth semantic analysis.First of all, our focus was on training and fne-tuning the LLM.We established a hierarchical classifcation for verbal aggression, encompassing three levels of impact: eliciting negative emotions (Level 1), inducing psychological imbalance (Level 2), and inficting psychological trauma (Level 3).The escalation in the severity of verbal aggression is delineated across these three levels.In addition, expressions containing apology phrases, such as "I'm sorry," fall within the category of language associated with apologies, and we annotate them as Level 4. It is important to note that in each interaction round, regardless of the severity of verbal aggression, the level 4 apology can only be initiated once.
Subsequently, the level of verbal aggression obtained from the LLM are seamlessly integrated into the control code.The development of the control code is implemented through Python programming, facilitating the conversion of emotional signals extracted from the LLM into digital signals suitable for Arduino.

Output
Arduino receives the previously mentioned digital signals that represent graded levels, corresponding to the infation time of the air pump, set to 3 seconds, 5 seconds, and 7 seconds, respectively.With increasing infation time, the expansion level of the airbag gradually rises.Upon receiving digital signal 4, Arduino promptly initiates the other pump for air extraction, with a duration of 2 or 3 seconds.

Tangible Representation of Verbal Aggression
Through the aforementioned apparatus, we map the three levels of verbal aggression to the infation levels of the airbags, further translating into tactile feedback in the form of sense of pressure experienced by the puppet wearer (Figure 3).When the degree of verbal aggression is "eliciting negative emotions," a 3-second infation period is triggered, resulting in level 1 infation.Consequently, the puppet wearer experiences mild discomfort.At this stage, expressing an apology, such as "I'm sorry," will restore the robot to its initial state.
As verbal aggression escalates to "inducing psychological imbalance, " the system dynamically adjusts the infation status, causing the airbag to expand to level 2. This prolonged infation directly impacts the pressure experienced by the puppet wearer, inducing a heightened sense of psychological imbalance, reaching discomfort level 2. When expressing an apology to the robot again, the infation status of the robot will weaken, but cannot fully return to the initial state.
When verbal aggression reaches the level of "inficting psychological trauma, " the system further extends the infation time, causing the airbag to expand to the highest level.This extreme infation directly translates into an intensifed discomfort experience for the puppet wearer, simulating the impact of psychological trauma.Apologizing to the robot results in a reduction of its infation status, yet it still cannot completely revert to the initial state.This implies that the harm caused by verbal aggression cannot be solely rectifed through apologies.
After a round of interaction concludes, pressing the restart button will restore the robot to its initial state.
Our trial results indicate that even the highest level of squeezing does not cause harm to the wearer.

APPLICATION SCENARIOS
We explore the diverse applications of Embo, a robot that not only functions as an educational tool to address child-directed verbal aggression but also adapts to various settings, demonstrating its potential as a proactive instrument for enhancing communication and fostering empathy in everyday life.

Educational Tool
In recent years, experiential teaching has been widely used in psychological health education in schools in China, allowing students to experience the psychological states of others through roleplaying and other methods [11].However, the psychological trauma caused by verbal aggressive is difcult to detect, and students' psychological conditions are covert and defensive [3].Educational tools are needed to help aggressors understand the harm.
Embo, as an educational robot (Figure 5.a), is designed for children aged 6 and above, as well as for parents.As wearers of Embo, individuals assume the role of the "aggressor," experiencing the robot's tactile feedback, which varies with the intensity of verbal aggression.Users can explore diferent levels of verbal aggression and receive corresponding squeezing feedback by pressing the Restart button, facilitating a deeper understanding of verbal aggression and its irreversible harm.
In a classroom setting, the teacher can introduce the concept of verbal aggression and invite students to participate in an activity.Each student wears Embo, playing the role of the "aggressor" and engaging in diferent levels of verbal aggression towards their Embo.When they utter ofensive language, Embo responds with a corresponding tactile feedback by squeezing their hands.Although students can attempt to mitigate the impact on Embo with an apology, some efects of verbal aggression could not be fully reversed, helping students to perceive the lasting impact of their words.
Embo is also benefcial in parent awareness training sessions.In these sessions, the teacher can elucidate the notion and harm of verbal aggression.In casual conversations, parents wearing Embo gradually recall disputes and their causes.Embo provides tactile feedback of pressure based on the levels of verbal aggression in the dialogue.This experience aims to make parents who are unaware of verbal aggression feel the harm and irreversible impact, thereby encouraging them to avoid using aggressive expressions when communicating with their children.
In these scenarios, Embo establishes an immersive educational environment, engaging children and parents in understanding verbal aggression.This interactive approach is instrumental in cultivating empathy and awareness, while promoting the adoption of healthier communication methods.

Potential Application Scenarios
Beyond the aforementioned scenarios, Embo's potential extends to various settings as depicted in Figure 5, such as psychological clinics and family environments, where it can act as an active agent in addressing and mitigating the issue of verbal aggression in our everyday life.The scenario is envisioned as follows: In a psychiatric clinic, patients with uncontrollable verbal aggression habits but wishing to improve their condition turn to the therapist; there is no cure for this problem, but the therapist gives the patient a robot called Embo, which is small enough to be carried around at all times, and the image of Embo itself is a reminder when the patient is about to engage in uncontrollable verbal aggression.
In the family, Embo is a mutual friend of the husband and wife.Whenever they start to have a verbal confict, one of them will put on Embo, and with the tactile feedback of Embo he will quickly calm down, and the tensions will be de-escalated by Embo in this way.Embo can also be a regulator in the family parent-child get-together scenario.Parents wear Embo to tutor their children's homework.Embo reminds them to maintain composure whenever their emotions are out of control, which creates a great parent-child bonding atmosphere.
As the application of Embo expands to more contexts, the shape of the robot may be adjusted accordingly to make it more motivated to be worn, and we will continue to develop it.

CONCLUSION AND FUTURE WORK
Verbal aggression poses signifcant challenges in child development, leading to enduring psychological efects on both targets and aggressors.This paper introduces Embo, a hand puppet robot designed to transform verbal aggression into tactile feedback, fostering empathy and understanding.We detail Embo's design methodology, present its initial prototype, and discuss its current classroom applications and potential future uses.Embo presently incorporates insights from verbal aggression literature and utilizes advanced language models to establish a semantic-based judgment system.However, it currently lacks the ability to analyze tonal variations and simulate the persistent state during prolonged exposure to verbal aggression.Future research will explore how enhanced robotic interactive feedback can represent the harmful efects of verbal aggression, including user tests to evaluate diferent interactive feedback mechanisms.