What are the body images of social robots (includes the slides)

The slides

The slides are to be presented at the The 19th IEEE International Conference on Advanced Robotics and Its Social Impacts (ARSO 2023) Berlin.


The following article is not a transcription of the talk, but my related thoughts about the study.

Remarks on the concepts

Body image is among the most strange terms that I’ve seen in psychology. People with different backgrounds come up with totally different things upon hearing it. Therefore, before moving to the topic, it’s the best to make sure we don’t have any misconceptions about body image.

Body image is the way a person perceives their own body, and the way they think others perceive it. That means, it refers the visuo-spatial aspects of one’s body as well as the semantic derivatives in a generalized sense (body image has two entries on Wikipedia, in terms of psychology and neuroscience respectively). In robotics, the term appears to extend even a third definition, which is quite blurred with body schema.

Author note: There are a lot of insightful articles in the field that revealed the potential links between body image and body schema, e.g. this study and Hoffmann et al. 2010. They definitely worth another dedicated post.

Traditionally, in psychology, the distinction between body image and body schema lies between the original domains. Simply put, body image originates from the visuo-spatial modality and body schema originates from the sensorimotor modality (De Vignemont 2010). The mixing between the concepts is not to be blamed because these modalities are themselves non-separable within our forebrains (Pitron & De Vignemont 2017).

It is worth mentioning that body image is specifically all about one’s own body. In our study, the empirical results never mentioned how a robot has to do with the robots' awareness about its own body. That part of the body images is not as obvious as the empirical results, but we can demonstrate that with in a very simple way. At the end of the presentation, I tried to simply prompt ChatGPT to imagine itself as a specific robot. Thanks to the (intelligent) monster, we saw how quickly a naĂŻve language model was into the role with just a few words! Moreover, recall how the body images were built by a group of people by freely expressing their first impressions towards the robots, the body images complied to our expectations!

In brief: an examination of social robots within a semantic space

The purpose of this study is to investigate how humans form semantic representations of robots based on visual characteristics. Visual input is transformed into semantic representations in the brain, and a robot’s visual characteristics includes not only its visual appearance but also its semantic meaning and social significance. Therefore, the visual characteristics of a robot encompasses not only the visual appearance of the robot but also its semantic meaning and social significance.

A word embedding model can accurately capture real-world semantic information by calculating the likelihood of words appearing next to each other in large, naturally occurring texts typically found on the internet using a single-layer neural network. To further improve the understanding of what a social robot is, we propose to embed social robots into such a semantic space. Rather than focusing on the visual appearance of robots, which can be difficult to quantify, we therefore characterize a robot by analyzing words that are spontaneously associated with the robot.

A free association task was used to collect basic impressions about various robots and to model these data within a semantic space using word vectors. The affect levels of the words were evaluated using “valence”, “arousal” and “dominance” measurements from an affect lexicon. Finally, we explored how the vector that characterize a given robot relate to the word “PERSON” within the semantic space, to assess how closely participants associate the robot with human environment and its psychological impacts. Taken together, the study demonstrates the importance of semantic representations in studying human-robot interaction.

References

  • Hoffmann, M., Marques, H., Arieta, A., Sumioka, H., Lungarella, M., & Pfeifer, R. (2010). Body schema in robotics: a review. IEEE Transactions on Autonomous Mental Development, 2(4), 304–324.
  • De Vignemont, Frederique (2010). Body schema and body image—pros and cons. Neuropsychologia, 48(3), 669–680.
  • Pitron, V., & de Vignemont, Frederique (2017). Beyond differences between the body schema and the body image: insights from body hallucinations. Consciousness and Cognition, 53(), 115–121.