Robotic active tactile sensing inspired by serotonergic modulation using active inference

Abstract

When faced with uncertainty in the world, biological agents actively sense the environment to acquire the most informative input to fulfil their tasks. Actions are performed to adjust bodily sensors to maximize the collected information, which is usually known as active sensing. For instance, rodents continuously adjust the speed and amplitude of whisking to better identify objects and body location in space, which ultimately regulates navigation. Whilst, the internal mechanism that drives active sensing in humans is still under research, recent evidence points towards neuromodulators, such as serotonin, that influence whether the habitual behaviour is preferred over sensor adjustments to trigger exploration. Here, we present an active tactile-sensing model for a robot inspired by the serotonergic function viewed from the uncertainty minimization perspective. To mechanistically explain this neuromodulatory function, we associated it with precision parameters regulating habitual behaviour and tactile encoding based on previous findings. We qualitatively evaluated the model using an experiment inspired by the gap-crossing paradigm but tailored to a humanoid with tactile sensing. Behavioural switch timing results show the strong dependencies between active sensing and precision regulation. Ultimately, this work discusses how the neural microcircuitry regulates active sensing, hence opening future research of such neuromodulatory processes translated to robotics active sensing and perception.