The Use of Social Sensorimotor Contingencies in Humanoid Robots
379
Social cognition is the ability to attribute meaning to the events that
unfold during social interaction and engage meaningfully with appropriate social
behaviour. Humans employ verbal and non-verbal communication channels
(gaze, proximate distance, gesture and facial cues [2]) to determine the social
meaning when interacting with another person. These sensorimotor channels of
communication heavily rely on the context of social interaction, as the same
non-verbal cues can have different meanings in different contexts. Humans can
naturally understand these multimodal cues with a high level of precision and
differentiation. Thus, one of the main challenges of Human-Robot Interaction
(HRI) is to approximate or even resemble and contextually understand the richness of multimodal sensorimotor channel information.
The behaviour of social robots needs to be transparent and interpretable to
be accepted by humans, and we can use recent theories on social cognition in
humans as a framework from which to develop robots with interpretable and
meaningful social behaviour. Theories suggest that social cognition is not representational, nor a symbolic process as previously accepted [3]. Instead, the
ability to determine social meaning is developed through experience and shaping of social sensorimotor contingencies. Thus, social cognition and social skills
are the mastery of sensorimotor information that has been learned in past experience in congruent social situations [4]. Sensorimotor contingencies represent
action-motor loops that enable the categorisation of events, actions, objects
and fluent interaction between an agent and its environment. This approach
toward non-verbal interactions is supported by both psychology and neuroscience findings [5]. When applied in robotics, sensorimotor contingencies can
be used, for example to study real-time collaboration and social entrainment in
dyadic or multi-agent interactions when agents reciprocally and actively attend
the actions of the partner [6]. Social behaviour and the communication channels
employed are important factors that affect bonding, empathy and attributing
mental states when interacting with a synthetic agent [7] and humans naturally
tend to anthropomorphise non-human interactions. Thus, social salience can be
utilised to modulate human expectations, and when expressed via social sensorimotor contingencies, it could be a way to improve the fluency and transparency
of social interactions.
This work is grounded in the principles of sensorimotor contingencies and
aims to investigate whether they can be a solid basis upon which to build a
model for interactions with humans. We wish to extend dyadic interactions to
group social activities, and study which sensorimotor contingencies are relevant
and what type of behaviour (personality) displayed by a synthetic agent best
takes advantage of humans’ natural tendency to attribute social salience. The
social activity we have chosen is that of taking a group “selfie”. To be able to
develop a synthetic agent that groups people together to take a selfie, we first
need to answer the following questions: ”How do humans perform this task using
sensorimotor contingencies?”, followed by “Are people capable of correctly reading the sensorimotor information displayed by a synthetic agent?”, and finally,
“Can the behaviour of a synthetic agent that relies only on sensorimotor infor-
379
Social cognition is the ability to attribute meaning to the events that
unfold during social interaction and engage meaningfully with appropriate social
behaviour. Humans employ verbal and non-verbal communication channels
(gaze, proximate distance, gesture and facial cues [2]) to determine the social
meaning when interacting with another person. These sensorimotor channels of
communication heavily rely on the context of social interaction, as the same
non-verbal cues can have different meanings in different contexts. Humans can
naturally understand these multimodal cues with a high level of precision and
differentiation. Thus, one of the main challenges of Human-Robot Interaction
(HRI) is to approximate or even resemble and contextually understand the richness of multimodal sensorimotor channel information.
The behaviour of social robots needs to be transparent and interpretable to
be accepted by humans, and we can use recent theories on social cognition in
humans as a framework from which to develop robots with interpretable and
meaningful social behaviour. Theories suggest that social cognition is not representational, nor a symbolic process as previously accepted [3]. Instead, the
ability to determine social meaning is developed through experience and shaping of social sensorimotor contingencies. Thus, social cognition and social skills
are the mastery of sensorimotor information that has been learned in past experience in congruent social situations [4]. Sensorimotor contingencies represent
action-motor loops that enable the categorisation of events, actions, objects
and fluent interaction between an agent and its environment. This approach
toward non-verbal interactions is supported by both psychology and neuroscience findings [5]. When applied in robotics, sensorimotor contingencies can
be used, for example to study real-time collaboration and social entrainment in
dyadic or multi-agent interactions when agents reciprocally and actively attend
the actions of the partner [6]. Social behaviour and the communication channels
employed are important factors that affect bonding, empathy and attributing
mental states when interacting with a synthetic agent [7] and humans naturally
tend to anthropomorphise non-human interactions. Thus, social salience can be
utilised to modulate human expectations, and when expressed via social sensorimotor contingencies, it could be a way to improve the fluency and transparency
of social interactions.
This work is grounded in the principles of sensorimotor contingencies and
aims to investigate whether they can be a solid basis upon which to build a
model for interactions with humans. We wish to extend dyadic interactions to
group social activities, and study which sensorimotor contingencies are relevant
and what type of behaviour (personality) displayed by a synthetic agent best
takes advantage of humans’ natural tendency to attribute social salience. The
social activity we have chosen is that of taking a group “selfie”. To be able to
develop a synthetic agent that groups people together to take a selfie, we first
need to answer the following questions: ”How do humans perform this task using
sensorimotor contingencies?”, followed by “Are people capable of correctly reading the sensorimotor information displayed by a synthetic agent?”, and finally,
“Can the behaviour of a synthetic agent that relies only on sensorimotor infor-