EAGER: Context-Aware Canine Intent Recognition and Response for Interactive Canine-Robot Teams

The goal of this EArly-concept Grant for Exploratory Research (EAGER) project is to establish effective two-way communication between dogs and robots, culminating with robot-canine teams that can jointly accomplish collaborative tasks impossible for either one alone. The project proceeds in three stages. First, the robot will learn to interpret the dog's posture, movement, and vocalizations. The robot will learn which canine behaviors indicate, for example, hunger, fatigue, or the location of a specified object. Second, the robot will learn to respond to these cues, for example, by providing the dog with food or water, or by grasping and retrieving a designated object. Finally, the dog will be trained to perceive the robot as a source of useful actions, thus enabling important new capabilities unique to dog-robot teams. For example, the team may be asked to retrieve an object marked with an odor that only the dog can perceive, but which can be physically manipulated only by the robot. These teams would be potentially transformative in applications such as hazardous material detection and localization, contraband interdiction, and disaster response, where a semi-autonomous dog-robot team would allow a human handler to remain safely out of danger. Dog-robot teams could also significantly exceed the ability of either robots or dogs alone to assist aging or impaired individuals in their homes with the activities of everyday life. This project seeks to enable robots that can react to and initiate desired behaviors with a canine partner. The envisioned methodology is a hybrid control paradigm integrating an offline-trained model-free component, for high-level task competence, with model-based control, to improve stability and constraint satisfaction during real-world deployment. The project will create computable models to allow robots to internally represent canine cognition and to predict the corresponding behavior. These models will be used to minimize the training required for effective operation by the team, and to quantify changes in the dog's perception of the robot over time. Experiments will provide new static and dynamic labeled datasets to augment existing resources, and will demonstrate the use of dog-robot teams to accomplish representative tasks under challenging conditions. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.