A Multidisciplinary Artificial Intelligence Model of an Affective Robot

Abstract

A multidisciplinary approach to a novel artificial intelligence system for an affective robot is presented in this paper. The general objective of the system is to develop a robotic system which strives to achieve a high level of emotional bond between humans and robot by exploring human love. Such a relationship is a contingent process of attraction, affection and attachment from humans towards robots, and the belief of the vice versa from robots to humans.

The advanced artificial intelligence of the system includes three modules, namely Probabilistic Love Assembly (PLA), based on the psychology of love, Artificial Endocrine System (AES), based on the physiology of love, and Affective State Transition (AST), based on emotions. The PLA module employs a Bayesian network to incorporate psychological parameters of affection in the robot. The AES module employs artificial emotional and biological hormones via a Dynamic Bayesian Network (DBN). The AST module uses a novel transition method for handling affective states of the robot. These three modules work together to manage emotional behaviours of the robot.

Keywords

human- robot interaction affective artificial intelligence and affective computing

1. Introduction

Research in robotics ranges from industrial service targeting, optimizing practical systems in terms of accuracy and speed, to social robots with the goal of communication and interaction with humans. When it comes to interactions between humans and robots, a sense of attachment to objects will become an issue, so this triggers the idea that bi-directional emotional interaction between humans and robots might be needed in the future. This study introduces a new generation of robots, with the ability to love and be loved by humans. We refer to this as “Lovotics”, which explores the concept and possibilities of human-robot relationships.

The Lovotics robot was designed and developed using several design theories for the hardware [1] and various novel algorithms for the software [2, 3]. The overall system structure of the Lovotics robot is presented in Figure 1. The system consists mainly of four input modules, which captures basic sensory data such as sound, vision, touch and acceleration. A separate pre-processing algorithm is designated for each input [4, 5], as the raw data needs to be prepared before being sent to the artificial intelligence module which computes the love probability, hormonal levels and affective states of the robot. The output generator is a separate layer which maps the internal state of the robot towards rational behaviour. Various behavioural patterns of the robot are represented using outputs generated by audio devices, motion devices and illumination devices. This paper explains the artificial intelligence module of the Lovotics robot.

Figure 1.

The overall structure of the Lovotics robot

The artificial intelligence of the robot employs probabilistic mathematical models for the formulation of love. An artificial endocrine system is implemented in the robot by imitating human endocrine functionalities.

Thus, the robot has the capability of experiencing complex and human-like biological and emotional states as governed by the artificial hormones within its system.

The robot goes through various affective states during the interaction with the user. It also builds a database of interacting users and keeps a record of the previous interactions and degree of love.

The first step in Lovotics is to develop a deep understanding of the physics, physiology and emotions of the human being in order to model this in the robot.

Even though various fields have proposed ideas about the role and function of love, the current understanding about love is still quite limited. Furthermore, developing an affection system similar to that of the human being presents considerable technological challenges.

The proposed novel advanced artificial intelligence system of Lovotics includes an Artificial Endocrine System, Probabilistic Love Assembly and Affective State Transition modules.

The psychological unit of the Lovotics artificial intelligence calculates the probabilistic parameters of love between humans and the robot. Various parameters, such as proximity, propinquity, repeated exposure, similarity, desirability, attachment, reciprocal liking, satisfaction, privacy, chronemics, attraction, form and mirroring, are taken into consideration.

The physiological unit of the Lovotics artificial intelligence employs an artificial endocrine system consisting of artificial emotional and biological hormones. Artificial emotional hormones include Dopamine, Serotonin, Endorphin and Oxytocin. For biological hormones Melatonin, Norepinephrine, Epinephrine, Orexin, Ghrelin and Leptin hormones are employed which modulate biological parameters such as blood glucose, body temperature and appetite.

A wealth of information about a person's emotions and state of mind can be drawn from facial expressions, voice, gesture, etc. The affective system of the robot analyses system inputs to generate suitable states and behaviours for the robot in real-time. The affective system is modelled as closely to the human being as possible in order to be an emotionally engaging system.

The Lovotics robot is an active participant in the communication process, which adjusts its internal hormonal levels and affective states depending on inputs and feedback from the human over the long-term.

Lovotics should be considered as a multidisciplinary research field utilizing fundamentals concepts from philosophy, psychology, biology, anthropology, neuroscience, social science, robotics, computer science, engineering and artificial intelligence. The proposed approach provides a system allowing the development of functionalities based on multidisciplinary research yielding a new form of love relationships.

2. Background

In human beings, emotions are not computed by a centralized neural system and operate at many time scales and at many behavioural levels [6]. Human love and social bonding employ a push-pull mechanism that activates reward and motivation pathways [7]. This mechanism overcomes social distance by deactivating networks used for critical social assessment and negative emotions, while it bonds individuals through the involvement of the reward circuitry, explaining the power of love to motivate and exhilarate [8].

Human empathy probably reflects admixtures of more primitive affective resonance or contagion mechanisms, melded with developmentally later-arriving emotion identification and theory of mind/perspective taking [9].

Inspired by human affection functionality, a multimodal AI system is proposed for Lovotics which considers both short and long-term effective parameters of affection.

2.1. Probabilistic Love Assembly

Through many years of investigating human love, researchers have discovered several factors that result in human love. Among them propinquity, proximity, repeated exposure, similarity, desirability, attachment, reciprocal liking, satisfaction, privacy, chronemics, attraction, form and mirroring have been noted as some of the main reasons for love [10-18]. These effective factors could be taken into account in the robot to develop a systematic method for appraising the level of love between a robot and a human.

Figure 2.

Schematic of the Probabilistic Love Assembly (PLA) module

2.2. Artificial Endocrine System

The natural endocrine system is a network of glands which works closely with the nervous system to secret hormones in the blood which affect the target cells to regulate homeostasis, metabolism and reproduction [19]. Hormones are chemical messengers that play a key role in the endocrine system to maintain homeostasis.

Different approaches to implement the artificial endocrine system for robotics have been proposed.

An adaptive neuro-endocrine system for robotic systems [20] presented an adaptive artificial neural-endocrine system that is capable of learning online and exploits environmental data to allow adaptive behaviour.

An artificial homeostatic system [21] presents an artificial homeostatic system devoted to the autonomous navigation of mobile robots, with emphasis on neuro-endocrine interactions. The artificial neuro-endocrine kinematics model for legged robot obstacle negotiation [22] provides insight into the possibilities afforded by a novel artificial neuro-kinematics network, constructed to primarily aid obstacle negotiation for a hexapod robot, MAX II. The challenges of designing nervous and endocrine systems in robots [23] is discussed in conceptual terms, taking into account the feasibility of designing a nervous system and an endocrine system in a robot, and reflecting upon the bionic issues associated with such highly complex automatons.

2.3. Affective State Transition

Different methodologies have been used for dealing with the internal state of the robot. State machine [24] and fuzzy state machine [25] are simple proposed methods for that purpose. A multiobjective evolutionary generation process for artificial creature's specific personalities was proposed, where the dimension of the personality model is defined as that of optimization objectives [26]. In that approach an artificial creature has its own genome in which each chromosome consists of many genes that contribute to defining its personality. As a biological approach, the finger blood pulse fluctuations are used for developing the online system to estimate a driver's internal state [27].

A large number of studies employed the OCC model [28] as the fundamental model of emotion. For example, it has been integrated for modelling the emotions in the embodied character [29] or to develop a distributed and computational model which offers an alternative approach to model the dynamic nature of different affective phenomena, such as emotions, moods and temperaments, to provide a flexible way of modelling their influence on the behaviour of synthetic autonomous agents [30].

TAME is a framework for affective robotic behaviour that deals with an exploratory experimental study to identify relevant affective phenomena to include into the framework in order to increase the ease and pleasantness of human-robot interaction. TAME stands for Traits,

Attitudes, Moods and Emotions, the four components of the personality and affect module that is responsible for producing affective behaviour [31, 32].

Machine intelligence needs to include emotional intelligence and demonstrate results toward the goal: developing a machine's ability to recognize the human affective state given four physiological signals and compare multiple algorithms for feature-based recognition of emotional state from this data [33].

Affective state transition can be designed as an emotional model. Energy and tension are known to be the two principal parameters for representing the emotions of a human being [34, 35]. Based on this fact, emotional categories can be mapped to arousal, valence and stance dimensions in a robot [36].

3. Methodology

3.1. Probabilistic Love Assembly (PLA)

The PLA module is employed to calculate the probabilistic parameters of love between a human and the robot. This module employs audio and touch as channels of interaction between the robot and humans as illustrated in Figure 2. The visual channel of the robot is employed for detection, recognition and expression analysis.

The following definitions are used for the PLA module:

Interactive zone of a robot is defined as an area in which a robot interacts with a human. It can be defined according to the physical properties of a robot e.g. size.

Interaction happens when a human is located inside the interactive zone of a robot.

Audio parameters include the number of harmonics, amplitude, pitch, tempo and envelope.

The following notations are also used:

u is the interactive radius of the robot according to the interactive zone.

C is the area of the environment.

${A r a}_{r b t}$ is the area of the robot's surface.

${A r a}_{t c h}$ is the area of the robot's surface which is touched by the human.

T is the time of considering the system for computation in the PLA module

$T_{i n t}$ is the total time that the robot and human are interacting either through audio or touch channels.

$T_{a u d}$ is the total time of human-robot interaction via audio channel.

$T_{t c h}$ is the total time of human-robot interaction via touch channel.

$T_{a s y}$ is the total time that asynchronous similarity is visible in the human and robot audio parameters.

Based on psychological studies and the study of nonverbal behaviours, probabilistic mathematical models for the 13 identified factors of love were formulated to provide an interpretation of the intimacy between humans and robots. Such mathematical models can be presented by a Bayesian network depicting the relationship between love and its causal factors.

1) Proximity: physical distance between humans plays an important role in their feelings for each other [12, 13]. The effects of physical distance were mapped over into audio, with the aim of using audio proximity to emulate the effects of physical distance. A human is within the proximity of the robot when the human is located inside the interactive zone of the robot. Hence, the probability of proximity is the chance of the human being located inside the interactive zone of the robot.

P (P r o x) = \frac{π u^{2}}{C}

(1)

2) Propinquity: propinquity represents the familiarity in terms of spending time with each other [10].

The robot and human spend time together if their distance is within the interactive zone and they are interacting.

P (P r o p) = \frac{T_{i n t}}{T}

(2)

2) Repeated Exposure: repeated exposure to a particular individual can increase the familiarity and hence the feeling of liking for that individual [12]. A human and a robot have exposure when a human moves from outside of the interactive zone of the robot to inside.

P (E x p o) = P (P r o x) (1 - P (P r o x)) = \frac{π u^{2}}{C} (1 - \frac{π u^{2}}{C})

(3)

3) Similarity: the degree of similarity between two people is directly related to the feeling of love [12, 10]. Similarity between a robot and a human can be calculated via comparing their audio parameters.

P (S i m l) = 1 - \frac{| {A u d i o}_{R o b o t} - {A u d i o}_{H u m a n} |}{{A u d i o}_{R a n g e}}

(4)

The difference between human audio parameters ( ${A u d i o}_{H u m a n}$ ) and robot audio parameters ( ${A u d i o}_{R o b o t}$ ) are found in Equation 4 and the probability of similarity within the audio range ( ${A u d i o}_{R a n g e}$ ) is calculated.

4) Desirability: the desirable characteristics of the other can automatically cause a strong attraction which can lead to cases of love at first sight [12, 10]. In order to find the desirability of a person for the robot, audio input from the human can be compared to predefined desired audio parameters of the robot.

P (D e s r) = 1 - \frac{| {A u d i o}_{H u m a n} - {A u d i o}_{D e s i r e} |}{{A u d i o}_{R a n g e}}

(5)

By comparing the actual ( ${A u d i o}_{H u m a n}$ ) and desired ( ${A u d i o}_{D e s i r e}$ ) audio parameters of the human within the audio range, probability of desirability could be calculated according to Equation 5.

5) Attachment: attachment represents the important emotional bonding between humans [11, 12, 17].

A human and robot are defined attached when the human stays within the interactive zone of the robot for a long time and the human touches the robot. Hence, the overall probability of attachment depends on both interaction ( ${i n t}_{A u d}$ ) and touch ( ${i n t}_{T c h}$ ).

\begin{matrix} P (A t c h) = P ({i n t}_{t c h} \cap {i n t}_{a u d}) = \\ P ({i n t}_{t c h}) P ({i n t}_{t c h} |{i n t}_{a u d}) = \frac{T_{t c h}}{T} \frac{T_{i n t}}{T} \end{matrix}

(6)

Figure 3.

The overview of the Bayesian network for Probabilistic Love Assembly (PLA) module

In the same manner, other probabilities can be calculated accordingly.

6) Reciprocal liking: being liked by another will increase the positive feelings toward the other, which will in turn influence the feeling of love [12, 10]. Since it is not possible for the robot to directly understand the love feeling of the human, the probability of reciprocal liking can be estimated indirectly via observing human behaviours. We assume that the human would touch the robot more if he/she likes the robot.

P (R e c p) = \frac{T_{t c h}}{T} \frac{{A r a}_{t c h}}{{A r a}_{r b t}}

(7)

7) Satisfaction: fulfilling the needs of the other will increase the satisfaction of him/her and hence lead to a positive feeling of love [12, 10]. The robot needs to be close to the human being and to be touched by the human, so effective parameters of proximity, repeated exposure and touch could affect the satisfaction of the robot.

\begin{matrix} P (S t f c) = \frac{π u^{2}}{C} \frac{π u^{2}}{C} (1 - \frac{π u^{2}}{C}) \frac{T_{t c h}}{T} \frac{{A r a}_{t c h}}{{A r a}_{r b t}} \\ = \frac{π^{2} u^{4}}{c^{2}} \frac{T_{t c h}}{T} \frac{{A r a}_{t c h}}{{A r a}_{r b t}} (1 - \frac{π u^{2}}{C}) \end{matrix}

(8)

8) Privacy: spending time together and having close interaction with each other can also improve love feelings [12]. The robot and human have private time together if their distance is within the interactive zone and they are interacting.

P (P r v c) = \frac{π u^{2}}{C} \frac{T_{i n t}}{T}

(9)

9) Chronemics: chronemics is the study of the use of time in nonverbal communication [18, 14, 16].

It can be modelled as a function of asynchronous similarity time, touch interaction time and audio interaction time.

P (C h r n) = \frac{T_{a s y}}{T} \frac{T_{t c h}}{T} \frac{T_{a u d}}{T}

(10)

Attraction: interpersonal attraction acts like a force drawing people together and influences the feeling of love [12]. Attraction between a human and a robot can be modelled as a function of time, force and area of touch:

P (A t r c) = \frac{T_{t c h}}{T} \frac{f}{F} \frac{{A r a}_{t c h}}{{A r a}_{r b t}}

(11)

10) Form: the form of interaction is also important in formulating love states [16]. Force and area of touch generate the form of interaction between the human and the robot.

P (F o r m) = \frac{f}{F} \frac{{A r a}_{t c h}}{{A r a}_{r b t}}

(12)

11) Mirroring: mirroring refers to matching the audio parameters of our robot with that of the person. It has been shown that a higher degree of synchronization between interactants is reflective of higher levels of interest between them [15]. Asynchronous audio variables of the robot and human could be compared to find similar properties at different times.

P (M i r r) = \frac{T_{a s y}}{T}

(13)

According to the presented probabilistic nature of the above 13 parameters, a Bayesian network is employed to link audio, touch and location to these effective parameters in order to estimate the long-term probability of love. The proposed Bayesian network is presented in Figure 3. This Bayesian network analyses various causal parameters of love between the robot and the human that can be categorized into three groups:

Human and robot distance and their interaction.

Synchronous and asynchronous audio parameters of the love between the human and the robot.

Duration, force and area of the touch.

The above variables can be correlated with their conditional independencies via a directed acyclic graph to generate a probabilistic model that consists of location, audio and touch variables as system inputs, 13 causes of love as intermediate events and overall love probability as the outcome of the Bayesian network.

Table 1.

Emotional hormones

Hormone	Effects	Virtual function
Dopamine	Excitement, Alertness	Activity
Serotonin	Happiness, Tension	Happiness
Endorphin	Sense of Well-Being	Contentment
Oxytocin	Trust, Empathy, Love	Love

Table 2.

Biological hormones

Hormone	Effects/Action	Virtual function
Melatonin	Drowsiness	Drowsy, Sleepy
Norepinephrine	Blood pressure	Excited, Active
Epinephrine	Blood Glucose	Agile
Orexin	Heart Beat Rate	Stimulated, Motivated
Ghrelin	Appetite	Hungry, Dizzy
Leptin	Metabolism	Lazy, Listless

3.2. Artificial Endocrine System (AES)

Hormones which are related to emotions and biological qualities according to the human endocrine system [37, 7] are utilized in the AES module of Lovotics.

Artificial emotional hormones, which include Dopamine, Serotonin, Endorphin and Oxytocin, are closely related to human emotions and presented in Table 1. Biological hormones Melatonin, Norepinephrine, Epinephrine, Orexin, Ghrelin and Leptin are employed to modulate biological parameters such as blood glucose, body temperature and appetite. Biological hormones and their effect on the AES module are presented in Table 2.

The level of emotional and biological hormones can be considered as internal variables of the artificial endocrine system. This system in nonlinear and can be expressed in the form of first-order differential equations as below:

\dot{x_{t}} = f (x_{t}, u_{t}), y_{t} = h (x_{t}, y_{t})

(14)

Where, $u \in R^{v_{1}}$ , v¹ is the number of input variables which includes touch area, touch force, number of harmonics, amplitude, pitch, tempo, envelope, battery charge, motor situation and network situation, $y \in R^{v_{2}}$ , v² is the number of output variables which includes tilt degree, vibration level, colour values, position parameters, speed parameters, rotation parameters and height, $x \in R^{v_{3}}$ , v³ is the number of endocrine state variables which includes Dopamine, Serotonin, Endorphin, Oxytocin, Melatonin, Norepinephrine, Epinephrine, Orexin, Ghrelin and Leptin, and f(.) and h(.) are vector-valued nonlinear functions.

The above nonlinear artificial endocrine system can be managed through the use of a Dynamic Bayesian Network (DBN) [38, 39]. There are four major layers in the DBN system, namely input, endocrine, emotional and behavioural layers. In the input layer, touch and audio are the two main preceptors with their respective parameters. In addition, internal situations of the robot such as battery life, motor conditions and networking situations are considered to regulate the biological hormones.

The endocrine layer consists of emotional and biological hormones. In the emotional layer six basic emotions and six moods are used to model the emotional situation. In the behavioural layer, tilt degree, vibration level, colour, 2D movement, 2D speed, rotation and height are output parameters of the robot. Lovotics DBN is illustrated in Figure 4.

3.3. Affective State Transition (AST)

The main two dimensions of emotion are considered as Activation ( $\vec{A c t}$ ) and Motivation ( $\vec{M o t})$ axes in the two dimensional affective state plane in order to develop an affective space. The third dimension of this affective space is also considered to represent Sub-States via the ( $\vec{S u b}$ ) axis. This third axis is used to present the rate of changes in the main two dimensions. Hence, the affective space of the robot is defined as a three dimensional Cartesian coordinate system with axes: [( $\vec{A c t}$ ), ( $\vec{M o t})$ , ( $\vec{S u b})$ ].

Using the above affective space, a novel transition system is proposed which could handle the immediate emotional properties of the robot. This short-term emotional module cooperates with the two other AI modules of the Lovotics robot to manage the overall internal state of the robot.

In order to model the system to link interaction and affective states, the transition in the affective state space is formulated as below:

{(\vec{S})}_{T_{A c t, M o t, S u b}} = {(\vec{S})}_{t - h} + η (\vec{Φ}) + β Γ \vec{Δ}

(15)

where ${\vec{S}}_{T}$ is the affective state of the robot in the affective space at time t and ${\vec{S}}_{t - h}$ is the affective state in time t − h and where h is the sampling time. $\vec{Φ}$ is the vector field over the states which converges to a stable point in the affective state coordinate system.

Figure 4.

Dynamic Bayesian Network of Lovotics

Vector $\vec{Φ}$ can be considered as the gravitational field of a point mass c, located at point $r_{1} \in R^{3}$ having position $r_{0} \in R^{3}$ as:

\vec{Φ} = \frac{- k c}{| {\vec{r}}_{-} {\vec{r_{0}}}_{} |} ({\vec{r}}_{-} {\vec{r_{0}}}_{})

(16)

where c, k ∈ℜ are constant numbers, $\vec{r}, \vec{r_{0}} \in R 3$ , $\vec{Φ}$ points toward the point $r_{0}$ and has magnitude |Φ| = $\frac{k c}{{| {\vec{r}}_{-} {\vec{r_{0}}}_{} |}^{2}}$ .

η is the adjusting parameter for the converging vector field.

β is the affective state coefficient which can be assigned to regulate the rate of changes in the affective state.

Γ is the learning rate.

$\vec{Δ} = {\vec{Δ}}_{A c t, M o t, S u b}$ , is the three dimensional normal vector to transfer the state over time in the affective state space based on the emotional input according to interactions.

The first two components are in the Activation – Motivation plane which are driven by emotional input:

{\vec{Δ}}_{A c t, M o t} = \sum_{m = 1}^{6} e_{{M o t}_{m}} \vec{M o t} + \sum_{m = 1}^{6} e_{{A c t}_{m}} \vec{A c t}

(17)

where $\vec{M o t}$ and $\vec{A c t}$ are representations of the Motivation and Activation axes in X and Y directions in a Cartesian coordinate system and vector e corresponds to the six values of the basic emotions, which are happiness, sadness, disgust, surprise, anger and fear, in Activation and Motivation directions.

The third component of $\vec{Δ}$ represents the movement in the sub-state direction which is obtained from the rate of the first two components:

{\vec{Δ}}_{S u b} = |\frac{d}{d t} {\vec{Δ}}_{A c t, M o t}| \vec{S u b}

(18)

Where $\vec{S u b}$ is the representation of the sub-states axes, in the Z direction of the Cartesian coordinate system. In this way vector βΓ $\vec{Δ}$ finds its direction to reach the next affective state.

Hence, the overall affective state formula, considering the interaction and transition methodology, can be presented as:

\begin{matrix} {\vec{S}}_{t_{(A c t, M o t, S u b)}} = \\ {\vec{S}}_{t - h} + η \frac{k c}{{|{\vec{r}}_{-} {\vec{r_{0}}}_{}|}^{2}} + β Γ (\sum_{m = 1}^{6} e_{{m o t}_{m}} \vec{M o t} + \\ \sum_{m = 1}^{6} e_{{a c t}_{m}} \vec{A c t} + |\frac{d}{d t} Δ_{A c t, M o t}| \vec{S u b}) \end{matrix}

(19)

In this way, the short-term affective state of the robot can be generated for computing the immediate sentimental properties of the robot. Using this short-term affective property, the robot is capable of undergoing several affective states by employing the Affective State Transition module which handles transformations from one affective state to another. As presented, these transitions are based on previous states, current mood and interaction influences from the environment.

3.3. AI fusion

The artificial module of Lovotics robot, fuses data from three elaborated AI modules to control the overall behaviour of the robot as presented in Figure 5. In practice these three AI modules are connected to the output layer via a neural network system.

Figure 5.

Fusion of the three AI modules of the Lovotics robot

3.4. Results

For initial testing of algorithms and to visualize the interactions of Lovotics, a simulator program is developed for Lovotics. This simulator includes a virtual model of the robot, humans and the corresponding interactions within the environment. The main interface of the simulator is presented in Figure 6.

In the Lovotics simulator a robot is placed between several humans and left to interact with them. During interaction in the environment, the levels of six biological and four emotional hormones change accordingly.

Furthermore, the probability of love between the robot and different interactants was being calculated dynamically in the PLA module as illustrated in the simulation environment in Figure 6.

Figure 6.

Lovotics simulator. Right: simulated model of the robot's environment. Left: (from top) emotional and biological hormones levels and the probability of love between the robot and humans in the environment.

Figure 7 shows how the affective state of the robot and the probability of love were changing during interaction.

Values of activation, motivation and sub-state were changing during the interaction, and the probability of love between the robot and certain humans was increased during two hours of interaction in the simulator.

Figure 7.

Changes in the AST (top) and PLA (bottom) modules of the robot during interaction in the Lovotics simulator

A robot was successfully designed and developed as an affective agent in social interactions. The robot is designed to inculcate love with humans, as well as express emotion via artificial nonverbal behaviour. An artificial intelligence system was developed to help realize the presentation methods of the robot and systematically implement the design into a robotics system. Figure 8 shows the design of the Lovotics robot.

Figure 8.

The hardware structure of the robot

The graphical user interface (GUI) of the Lovotics software is presented in Figure 9. The Lovotics GUI is also accessible via a remote connection while the robot is in operation.

The left part of the GUI represents values of input parameters after being evaluated by each pre-processor module. Data from the accelerometer for localization and touch sensors for tactile input is presented first, then the interactant's face is detected and recognized with confidence level and audio input parameters such as pitch. The middle part of the GUI represents the PLA and AST values of the AI modules.

Figure 9.

Graphical user interface of Lovotics

The 13 related parameters of the PLA module and probability of love with the detected person is presented. The affective state parameters are also presented in this section. The right part of the GUI serves to display the output of the robot, which reflects the current behaviours and emotion of the robot, including RGB data of LEDs, omni-wheel movements and servo motors for deforming the outer cover. Values of hormones in AES module are also presented in this part.

The Lovotics robot is shown during interaction in Figure 10. Videos of the robot are available at the Lovotics website (www.lovotics.com).

4. Conclusion

The novel advanced artificial intelligence system of Lovotics was presented which includes three modules, namely the Artificial Endocrine System (AES), which is generally based on the physiology of love, the Probabilistic Love Assembly (PLA), which is based on the psychology of love and Affective State Transition (AST), which is based on emotions. Inspired by the science of human love, these modules were designed and developed by employing artificial intelligence methods like Bayesian networks. These three modules work closely together in order to generate realistic behaviours by the robot.

By employing this artificial intelligent system, we have developed an intelligent robotics system, capable of emulating the love process of a human being, to establish long and meaningful relationships with humans. This holds potential for creating robots that can form meaningful bi-directional bonds with humans based on a level of emotional intelligence that is still lacking from robots today.

Figure 10.

Lovotics robot during interaction

Lovotics could lead to a revolutionary love relationship between humans and robots, via real emotive actions such as kissing or hugging. In the future, these more intimate and close interactions can be incorporated in the robot to further enhance the experience of love between humans and robots.

Lovotics introduces a novel interactive robot, with a focus on inculcating intimacy with humans via a slow communication process. Through long-term interaction with the robot, interactants build up an emotional attachment with the robot. This could create a new form of bi-directional relationship between humans and robots.

The aim of this system is to pave the way to creating personal relationships between humans and robots in the form of Lovotics, to exhibit love between humans and robotics. Lovotics introduces an interactive method which focuses on inculcating intimacy with humans.

This holds potential for creating robots that can interact with humans at a level of emotional sophistication that is still lacking from robots today.

Users experience responses based not only on immediate inputs, but also on prior interactions and relationships with particular humans. Constant close interaction with the robot can lead to a more intimate relationship. For example, if a new user strokes the robot, it may lead to a lower level of reciprocation when compared to similar action by a person familiar to the robot. Interactions are also influenced by parameters derived from the robot's internal state.

The next step is to conduct extensive qualitative fieldwork to help gather more data about user responses to the present developed robot and compare these with other forms of robots. Using this information, it will be possible to further redesign the robot to better evoke the sense of nurture and care within the robot and humans, thus further developing their love relationship.

Along with the development of the robot itself, we must try to understand and define the role of this new genre of robots in the social and cultural context. It is difficult to predict what the relationship will be between humans and robots in the future, and tough questions such as ‘How does this change our definition of love?’ or ‘Is this kind of love appropriate?’ need to be addressed. The way to do this is to continue this study via an exploration of ‘love’ and study how man and machine are evolving new identities and relationships, and in this, create a range of Lovotics robots to tackle these issues.

Footnotes

5. Acknowledgments

This research is carried out under CUTE Project No. WBS R-705-000-100-279 partially funded by a grant from the National Research Foundation (NRF) administered by the Media Development Authority (MDA) of Singapore.

References

Samani

H. A.

Cheok

A. D.

Tharakan

Koh

Fernando

, “A design process for Lovotics,” Human-Robot Personal Relationships, pp. 118–125, 2011.

Samani

H. A.

Cheok

A. D.

, “Probability of love between robots and humans,” in Intelligent Robots and Systems (IROS), 2010 IEEE/RSJ International Conference on. IEEE, 2010, pp. 5288–5293.

Samani

H. A.

Cheok

A. D.

Ngiap

Nagpal

Qiu

, “Towards a formulation of love in human-robot interaction,” in RO-MAN, 2010 IEEE. IEEE, 2010, pp. 94–99.s

Samani

H. A.

Cheok

Fernando

, “An affective interactive audio interface for Lovotics,” Computers in Entertainment (CIE), vol. 9, no. 2, p. 6, 2011.

Samani

H. A.

Ong

Hang

, “Active affective facial analysis for human-robot interaction,” in Robot and Human Interactive Communication, 2008. RO-MAN 2008. The 17th IEEE International Symposium on. IEEE, 2008, pp. 83–88.

Arbib

Fellous

, “Emotions: From brain to robot,” Trends in Cognitive Sciences, vol. 8, no. 12, pp. 554–561, 2004.

Esch

Stefano

, “The neurobiology of love,” Neuroendocrinology Letters, vol. 26, no. 3, pp. 175–192, 2005.

Zeki

, “The neurobiology of love,” FEBS letters, vol. 581, no. 14, pp. 2575–2579, 2007.

Watt

, “Social bonds and the nature of empathy,” Journal of Consciousness Studies, 12, vol. 8, no. 10, pp. 185–209, 2005.

10.

Aron

Dutton

Aron

Iverson

, Experiences of falling in love, Journal of Social and Personal Relationships, vol. 6, no. 3, p. 243, 1989.

11.

Tucker

Anders

, “Adult attachment style and nonverbal closeness in dating couples,” Journal of Nonverbal Behavior, vol. 22, no. 2, pp. 109–124, 1998.

12.

Levy

, Love + Sex with Robots: The Evolution of Human-robot Relations. New York, NY, Harper, 2007.

13.

Hall

, The hidden dimension: Man's use of space in public and private. Bodley Head, 1969.

14.

Wallbott

Scherer

, “Cues and channels in emotion recognition,” Journal of Personality and Social Psychology, vol. 51, no. 4, pp. 690–699, 1986.

15.

Grammer

Kruck

Magnusson

, “The courtship dance: Patterns of nonverbal synchronization in opposite-sex encounters,” Journal of Nonverbal Behavior, vol. 22, no. 1, pp. 3–29, 1998.

16.

Floyd

, “All touches are not created equal: Effects of form and duration on observers' interpretations of an embrace,” Journal of Nonverbal Behavior, vol. 23, no. 4, pp. 283–299, 1999.

17.

Brennan

Shaver

, “Dimensions of adult attachment, affect regulation, and romantic relationship functioning,” Personality and Social Psychology Bulletin, vol. 21, no. 3, p. 267, 1995.

18.

Bruneau

T. J.

, Chronemics and the Verbal-Nonverbal Interface, in Key

M. R.

, ed., The Relationship of Verbal and Nonverbal Communication, Walter de Gruyter, 1980.

19.

Hinson

Raven

Chew

, The Endocrine System. Elsevier Health Sciences, Edinburgh, 2007.

20.

Timmis

Neal

Thorniley

, “An adaptive neuroendocrine system for robotic systems,” in Robotic Intelligence in Informationally Structured Space, 2009. RIISS'09. IEEE Workshop on. IEEE, 2009, pp. 129–136.

21.

Moioli

Vargas

Von Zuben

Husbands

, “Towards the evolution of an artificial homeostatic system,” in Evolutionary Computation, 2008. CEC 2008. (IEEE World Congress on Computational Intelligence). IEEE Congress on. IEEE, 2008, pp. 4023–4030.

22.

Henley

Barnes

, “An artificial neuro-endocrine kinematics model for legged robot obstacle negotiation,” in 8th ESA Workshop on Advanced Space Technologies for Robotics and Automation, 2004.

23.

Cordova

Canete

, “The challenge of designing nervous and endocrine systems in robots,” in Management and control of production and logistics 2004 (MCPL 2004): A proceedings volume from the IFAC/IEEE/ACCA conference, Santiago, Chile, 3–5 November 2004. Elsevier, 2005, p. 321.

24.

Schulte

Rosenberg

Thrun

, Spontaneous, short-term interaction with mobile robots,” in Robotics and Automation, 1999. Proceedings. 1999 IEEE International Conference on, vol. 1. IEEE, 1999, pp. 658–663.

25.

Nourbakhsh

Bobenage

Grange

Lutz

Meyer

Soto

, “An affective mobile robot educator with a full-time job,” Artificial Intelligence, vol. 114, no. 1–2, pp. 95–124, 1999.

26.

Kim

Lee

, “Multi-objective evolutionary for specific personalities of artificial creature,” Computational Intelligence Magazine, IEEE, vol. 3, no. 1, pp. 43–53, 2008.

27.

Hashimoto

Katayama

, “Data processing method of finger blood pulse for estimating human internal states,” in Sae Publication Sp-1358. Human Factors In Driving, Vehicle Seating, and Rear Vision: Papers Presented at the 1998 Sae International Congress & Exposition, February 23–26, 1998, Detroit, Michigan, Usa (Sae Technical Paper 980016), 1998.

28.

Ortony

Clore

Collins

, The cognitive structure of emotions. Cambridge Univ Pr, 1990.

29.

Bartneck

, “Integrating the occ model of emotions in embodied characters,” 2002.

30.

Velásquez

, “Modeling emotions and other motivations in synthetic agents,” in Proceedings of the National Conference on Artificial Intelligence. Citeseer, 1997, pp. 10–15.

31.

Moshkina

Arkin

, “Human perspective on affective robotic behavior: A longitudinal study,” in Intelligent Robots and Systems, 2005.(IROS 2005). 2005 IEEE/RSJ International Conference on. IEEE, 2005, pp. 1444–1451.

32.

Moshkina

Arkin

R. C.

, “On TAMEing robots,” in SMC'03 Conference Proceedings. 2003 IEEE International Conference on Systems, Man and Cybernetics. Conference Theme - System Security and Assurance (Cat. No.03CH37483), 2003, vol. 4, pp. 3949–3959

33.

Picard

Vyzas

Healey

, “Toward machine emotional intelligence: Analysis of affective physiological state,” IEEE transactions on pattern analysis and machine intelligence, pp. 1175–1191, 2001.

34.

Russell

, “A circumplex model of affect,” Journal of Personality and Social Psychology, vol. 39, no. 6, pp. 1161–1178, 1980.

35.

Thayer

, The biopsychology of mood and arousal. Oxford University Press, USA, 1989.

36.

Breazeal

, “Emotion and sociable humanoid robots,” International Journal of Human-Computer Studies, vol. 59, no. 1–2, pp. 119–155, 2003.

37.

Pfaff

, “Hormones and social behavior,” Recherche, vol. 67, p. 02, 2008.

38.

Murphy

, “Dynamic bayesian networks: Representation, inference and learning,” Ph.D. dissertation, University of California, 2002.

39.

Ghahramani

, “Learning dynamic Bayesian networks,” Lecture Notes in Computer Science, vol. 1387, pp. 168–197, 1998.