Cognitive Modeling in Robot Learning for Adaptive Human-Robot Interactions

We propose an interdisciplinary workshop focusing on the aspects of human-centered design from the lens of cognitive science. In this workshop, we discuss how cognitive modeling can be applied to promote robot learning for adaptive human-robot interaction (HRI) from diverse theoretical perspectives and research domains by bringing together speakers from engineering, human-machine interaction research and cognitive science. 

HRI can be implemented in different ways—either the human and the robot can collaborate in close contact (e.g., coordinated lifting tasks), or the human and the robot can interact remotely (e.g., teleoperation of a robotic system with the help of computer applications such as virtual reality), where the robot assists the human in tasks deemed too dangerous for direct human involvement or in tasks in hard-to-reach places or hostile environments. Both types of HRI require the human agent and the robotic system to adapt to each other and also with the interaction environment. For example, in a coordinated lifting task, the robot needs to learn and infer the human’s intentions or strategies (e.g., through interactive forces), predict the human’s actions, and react accordingly given the environment. 

Recent work in cognitive science and computational modeling can inform adaptive HRI for robotics in approaching the goal of value alignment. For example, forward generative models of human planning and decision making can help make predictions of human choices and behaviors given the human agent’s goal and the current state. Inverse models can help the robotic agent learn the human’s intentions, goals, or beliefs given observations of human behaviors and decisions.

Part of learning about the human’s goals and beliefs can be captured by learning about the human’s subjective utility function. Therefore, we believe that robot learning can also benefit greatly from the theoretical framework of computational rationality as a theory of human-machine interaction. This framework assumes that the human agent behaves in a way that maximizes expected utility, given cognitive and environmental bounds. More specifically, in order to provide better support for humans, while learning the human’s goals and beliefs, it is also important for the robot to learn and adapt to the human’s information processing bounds, physical limitations, and the interaction environment. Such considerations of the human agent will be more likely to reduce the inconvenience, threat, annoyance, or harm to human users, and provide further accessibility, functionality, and protection instead. This approach to improve the quality of interactions between humans and the robots is in line with the goal of  bringing a human-centered design approach in the development processes for robotic systems. 

Given this background and motivation, our workshop focuses on three main topics: 

  1. How can human-centered design improve human-robot interactions?
  2. How can cognitive models inform robot learning? 
  3. Why is adaptive human-robot learning important and how can we model adaptive human-robot interaction? 

We believe that this workshop will be interesting for researchers from various related fields such as engineering, human-machine interaction and cognitive science. This workshop can provide an opportunity for researchers working in robotics and HRI from different perspectives to communicate with, understand, learn from each other, and ideally also find common grounds for future interdisciplinary collaborations with each other.

Invited Speakers:

  1. Dr. Anca Dragan (confirmed, in-person),
    Associate Professor, Electrical Engineering and Computer Sciences, UC Berkeley, USA.
    Topic: Cognitive Models Beyond Noisy-Rationality
  2. Dr. Patrick van der Smagt (confirmed, in-person/virtual),
    Director of AI Research, Volkswagen Group Chairman of Assistenzrobotik e.V.
    Topic: The latent Brain
  3. Dr. Henny Admoni (confirmed, virtual),
    Assistant Professor, Human And Robot Partners Lab, Robotics Institute, Carnegie Mellon University, USA
    Topic: Modeling Intent Through Nonverbal Behaviors in Human-Robot Interaction
  4. Dr. Andrew Howes (confirmed, virtual),
    Professor & Head of School, School of Computer Science, University of Birmingham, UK
    Topic: Models of Human Cognition for Artificial Intelligence
  5. Dr. Yukie Nagai (Confirmed, in-person),
    Project Professor, University of Tokyo
    Topic: Predictive Coding as a Unified Theory for Human and Robot Cognition


Organizers: Anany Dwivedi, Chenxu Hao

The workshop is proposed for the Conference on Robot Learning (CoRL), 2022 at Auckland, New Zealand.