Iterative Repair of Social Robot Programs from Implicit User Feedback via Bayesian Inference

04 Sep 2020 - Arun Balajiee

Talk Author: Michael Jae-Yoon Chung

Talk Date: 09/04/2020

Westworld is a TV series that talks about the idea of statisfying the needs of guests based on personalized programmed robots. The author of this paper, Iteartive Repair of Social Robot Programs from Implicit User Feedback via Bayesian Inference and this talk, quotes this TV series to be among his several inspirations behind the implementation. Social robotics is the field of interactions and dialogue between humans and robot. In this field, it has been a constant challenge to be able to personalize the robot to the individual interacting with the robot. Providing API or exposing low-level parameters for the end-user to program the social robot is not a feasible solution. Considering this the author talks about the idea of introducing a programming sketch language, Social Robotic Transition Sketch (SoRTSketch) Language, in order to write expressive programs. The language presents representations of Finte State Machines (FSM) for the correction variables of the interaction and a repair function to redistribute probabilities of the actions of the social robot based on implicit feedback from the user or interaction trace from the programmer. For example, the Finite State Machine has hole variables to allow the exact amount of wait time for the social robot, performing an action, required to wait when a user disengages interaction with the social robot and repairs based on this implicit feedback of disengagement from the user or an update to the function from the programmer. The errors in Finite State Machine transition happens because of incorrect transitions to perform actions or missed transitions while perfoming actions. These errors are treated as feedback to be “repaired” in the system. The Bayesian Inference of these errors is to refer to the redistribution of the probability of the actions of the robot. In their results, they observed that the general feeling of the human participants’ interactions with the robot was positive – the participant took it as their responsibility to make the human-robot interaction better and they trusted the robot more.

We think this exploration and introduction of a new language push the frontiers in the field of social robotics. Providing convenient controls for users to interact with a social robot increases the possibility of human-robot interactions more common in the modern times.