Intent Recognition Through Goal Mirroring (Doctoral Consortium)

Abstract

As the human population and life expectancy increases so does the need for integrating robots and virtual agents closely into everyday human life. These agents may provide care and attention where otherwise the manpower is lacking. In order to create better agents, that interact seamlessly with humans we need to draw lessons from what we know of human social cognition. Designing an agent inspired by these processes will provide an agent that is more predictable, less-threatening and overall welcome to its’ human benefactor. One important aspect of human social cognition is the innate ability to perform quick and efficient intention recognition. This ability enables humans to reason about the hidden goals of other agents around them through observations of their actions. In humans this ability is hypothesized to come from the existence of a mirror neuron system. Mirror neurons have first been discovered in the early 90’s. These neurons were seen to fire both when a monkey manipulated an object and also when it saw another animal manipulate an object. Recent neuro-imaging data indicates that the adult human brain is also endowed with a mirror neuron system for matching the observation and execution of actions within the adult human brain [4, 6]. This system is hypothesized to give humans the ability to infer the intentions leading to an observed action using their own internal mechanism. It is also attributed to other high level cognitive functions such as imitation, action understanding, intention and language evolution. Consequently, the human mirror neuron system may be viewed as a part of the brains’ very own plan/goal recognition module. Inspired by mirroring processes we have developed Goal Mirroring. A fast, online method that works in continuous domains. Goal Mirroring uses a planner to dynamically generate plans for given goals, eliminating the need for the traditional plan library. In this we also build on previous approaches —plan recognition by planning (PRP)— [3]. However, while existing PRP based recognizers only operate in discrete domains and in an offline manner Goal Mirroring provides an efficient online PRP approach while operating in continuous domains. We have extensively evaluated this approach, over hundreds of experiments, while measuring

Cite this paper

@inproceedings{Vered2017IntentRT, title={Intent Recognition Through Goal Mirroring (Doctoral Consortium)}, author={Mor Vered}, year={2017} }