Robot self/other distinction: active inference meets neural networks learning in a mirror

“The optimizer used was Adam [18] and the training used batches of 200 samples and 1000 epochs.”

“The robot approximates the belief of his body by means of minimizing the free energy [10], computing the most plausible solution of his arm location using the error between the predicted sensory input and the observed one.”

“Defining the brain as an inference machine [19] that is able to interpret the world from partial information [9, 15], body perception is presented here as an unconscious process that adapts empirical information to current learned models.”

“Under the predictive coding [30] approach, the robot is able to learn (approximate) generative models of the world, especially the effects of the body in the world.”

“Humans perceive their body in a flexible manner [5] but maintaining their identity [12].”

“Humans perceive their body in a flexible manner [5] but maintaining their identity [12].”

“Defining the brain as an inference machine [19] that is able to interpret the world from partial information [9, 15], body perception is presented here as an unconscious process that adapts empirical information to current learned models.”

“In essence, we propose the “double comparator” model inspired by the work of [36].”

“Defining the brain as an inference machine [19] that is able to interpret the world from partial information [9, 15], body perception is presented here as an unconscious process that adapts empirical information to current learned models.”

“According to [4], humans can learn and predict the consequence of actions and these forward models are used to determine the source of the sensory events.”

“Based on the active inference paradigm [11,27], we formalized body perception and action as an optimization problem that minimizes the prediction error, i.e., the error between the expected sensory effect and the observed one [10].”

“Around the age of two, the majority of infants are able to pass the self-recognition mirror test [31], like ele-phants, dolphins and some non-human primates [2].”

“The same model can generalize to forward model learning using more complex probabilistic deep models [7] within the free energy framework.”

“Within the self-perception cognitive science literature it is common knowledge the comparator model [8].”

“However, it can be treated in two different levels [35]: the sensorimotor [25] and the cognitive [2].”

“The agent encodes its belief of its body µ with the following density [6]: Where the first two likelihoods are the sensor modality contributions (assumed independent between each other) and the third term is the prior density for the first-order dynamics, which also depends on the causal variables ρ .”

“Previous works on artificial self-recognition, such as [14,23,34], have been hardly criticized [1]: “Merely simulating certain features of self-recognition through training/programming does not mean that the underlying mechanisms are the same, similar, or even remotely related”.”

“However, it can be treated in two different levels [35]: the sensorimotor [25] and the cognitive [2].”

“Conversely to previous works on active inference we enabled the robot to learn those models taking inspiration from our latest contributions on body perception with function learning [21, 22].”

“However, it can be treated in two different levels [35]: the sensorimotor [25] and the cognitive [2].”

“Further work will focus on scaling to raw images [32] and including appearance cues.”

“On the other hand, contingency is possible to be learnt with spiking neural networks [29] and self/other distinction can be developed by gradually increasing spatiotemporal resolution [26].”

“This experiment has been argued to be hardly intertwined with self-awareness and social understanding [14] and thus consciousness.”

“Inspired by [17], the probability of being itself is computed by normalization and smoothing as follows: where K is a smoothing constant working as a Kalman gain.”

“Previous works on artificial self-recognition, such as [14,23,34], have been hardly criticized [1]: “Merely simulating certain features of self-recognition through training/programming does not mean that the underlying mechanisms are the same, similar, or even remotely related”.”

“This approach determines that humans are aware of producing an effect in the world through their actions (i.e. agency [16]) by means of the prediction error congruence.”

“Previous works on artificial self-recognition, such as [14,23,34], have been hardly criticized [1]: “Merely simulating certain features of self-recognition through training/programming does not mean that the underlying mechanisms are the same, similar, or even remotely related”.”

“This work presents a novel active inference computational model that combines free energy optimization from [27] with function learning from [22].”

“Some researchers criticized the simplistic vision of the model and proposed that contingency detection and causal inference should be also involved as processes on top of congruence [37].”

“Artificial intelligence approaches for self-recognition are quite far from the biological plausibility [13,24].”

“Artificial intelligence approaches for self-recognition are quite far from the biological plausibility [13,24].”

“On the other hand, contingency is possible to be learnt with spiking neural networks [29] and self/other distinction can be developed by gradually increasing spatiotemporal resolution [26].”

“Lanillos et al. (2020) extend the active inference implementation presented in Oliver et al. (2019).”

“This embodiment [49] is crucial for social robots, such as personal aides, auxiliary robot nurses and companions, e.”

“…2020; Fountas et al., 2020; Mazzaglia et al., 2021; Millidge, 2020; Tschantz et al., 2019; Tschantz, Millidge, et al., 2020), and robotics (Çatal et al., 2021; Lanillos et al., 2020; Oliver et al., 2021; Pezzato et al., 2020; Pio-Lopez et al., 2016; Sancaktar et al., 2020; Schneider et al., 2022).”

“availability of generative processes that predict the effects of our body in the world [24,44].”

“More specifically, the comparator model has been criticized as lacking important top-down components, i.e. sensorimotor contingency detection and causal inference, that are important for a more realistic model of self-detection (Lanillos et al., 2020b).”

“ness [1], [2], and multisensory integration [17], [20].”

“These elements have been studied as constructive approaches in cognitive developmental robotics [93,94].”

“Robotics research has begun to propose models to achieve self-other distinction in robots (e.g., [21], [22]).”

“Notably, even AI systems that can reason about the activities of other agents 27 and develop a self-other distinction 28 .”

“In this section, we explain and discuss the following approaches: Sancaktar et al. (2020), Lanillos et al. (2020), and Oliver et al. (2019).”

“In the last years, some proof-of-concept studies in robotics have shown that AIF may be a powerful framework to address key challenges in robotics, such as adaptation, robustness and abstraction in goal-directed tasks [5], [6], [7], [8].”

“In the latter, we can exploit the backpropagation method to compute it (Lanillos et al., 2020b).”

“In robotic self-recognition tasks, researchers typically use body actions, such as robotic manipulators, and visual cues to enable humanoid robots to determine whether the motion changes appearing in a mirror are the result of their movements in the world [6], [7].”

“Differently, in some neurorobotics models, forward model plays the core role in high-level cognitive functions such as self/other distinction, sense of agency or body ownership [37]–[39].”

“agents’ body (in forms of generative images or joint estimation), other models like (Lang et al., 2018; Hinz et al., 2018; Lanillos et al., 2020) enable agents to distinguish”