by Justin W. Hart, Richard G. Freedman, Nick DePalma, Luca Iocchi, Matteo Leonetti, Emmanuel Senft, Elin A. Topp, Ross Mead
The AAAI symposium on “Artificial Intelligence for Human-Robot Interaction (AI-HRI): Service Robots in Human Environments” was held at the Westin Arlington Gateway, Arlington, Virginia from November 7- 9, 2019. This is the sixth AI-HRI Fall Symposium, bringing together researchers whose work spans areas contributing to the development of human-interactive autonomous robots. In the past few years, technologies related to or deployed on service robots has become a popular research topic in this area. This year’s theme invited researchers to look at problems and frame discussion through this lens.
Service robots are (semi-)autonomous robots that assist humans in perform- ing tasks. For service robots to be useful, they must gracefully interact with humans in homes, workspaces, and public places over long periods of time, doing useful work while also demanding little supervision. The area of service robots has seen rapid growth over the past few years in research, in academia, and as applied in real-world settings. RoboCup@Home, the World Robot Summit Challenge, the European Robotics League, and other similar competitions have focused on service robots that perform tasks in public spaces, such as cleaning homes or serving drinks at a party or in a coffee shop. Companies and research entities such as Diligent Robotics, Savioke, and the Toyota Research Institute6 are in various stages of deploying robots in hospitals, hotels, and homes. The pursuit of service robots has led to a proliferation of exciting new research problems that bridge the gap from the laboratory to the real world.
Twenty-seven papers were presented at this year’s AI-HRI symposium by participants from universities, industry, and national research laboratories. Topics included service robot applications (from shopping mall guides to rehabil- itation coaches to selfie-taking robots), communication through a variety of interface modalities (from physical devices to natural language to gestures to augmented reality), multi-agent teaming and collaboration, cognitive modeling of interaction partners, dataset features and generation, and social evaluations of humans observing or interacting with robots, among others. In addition to the paper presentations, invited talks were given by Michael Gleicher (University of Wisconsin-Madison), Manuela Veloso (Carnegie Mellon University and J.P. Morgan AI Research), Laura Hiatt (Naval Research Laboratory), and Matthew Marge (Naval Research Laboratory). Attendees also participated in multiple breakout discussions that highlighted many relevant questions and potential di- rections regarding the near future of service robots. This provided a series of guidelines to prepare service robots for interaction with people around them in the real world and opportunities to shape how the AI-HRI community should ensure that these guidelines can be incorporated.
Several key themes emerged through the breakout discussions, referencing points from the presented works and invited talks. Inspired by discussion of Manuela Veloso’s talk, one recurring theme of interest was the concern that AI-HRI researchers are developing systems in a “scientific vacuum”. Limiting robots to interacting with people in laboratory spaces and controlled environments is likely to miss important challenges as they relate to real-world systems. However, the participants saw many challenges in overcoming this issue due to the lack of incentives for researchers to leave this vacuum: cheaper hardware would allow greater risks if the robot broke, acceptance of publications that discuss failure would invite studies of complex systems that may fail to validate an initially-held hypothesis, and a research culture allowing replication stud- ies would encourage scientists to explore how others’ work (does not) apply in their local environments. These incentives would encourage the development of more resilient systems that could be used for long-term situations rather than one-shot experiments. Another ongoing discussion throughout the symposium concluded that while it is highly likely that we will see a variety of service robots deployed in the next ten years, driven by a demand for practical work to be accomplished, it is unlikely that people will yet engage in long-term so- cial interactions with household robots within that timeframe. This is due to a combination of open research questions regarding “natural” and “intuitive” HRI, as well as how to foster the penetration of technology into society and improve the social acceptance of robots. Lastly, there was a lively discussion of the limitations of current robotics tools, libraries, and datasets. There are many features that we have a practical knowledge of how to implement in these tools that simply are not broadly supported. The community has access to a wide range of tools, but integrating them can be complicated. Additionally, as these tools are often supported by single individuals, optimization might be missing. This performance limitation prevents the application in the real world, especially when integrated in a complex architecture. As a community effort, it was proposed that we could improve the basic, widely-used tools of the trade to help fuel long-term innovation.
Overall, AI-HRI 2019 was a very productive and stimulating symposium, and the attendees are excited to continue working towards research that will prepare both robots and people to more intuitively interact with each other in the future!
Justin W. Hart, Nick DePalma, Richard G. Freedman, Luca Iocchi, Matteo Leonetti, Katrin Lohan, Ross Mead, Emmanuel Senft, Jivko Sinapov, Elin A. Topp, and Tom Williams served as the program committee of this symposium. The peer-reviewed papers of the symposium were published on arXiv at https: //arxiv.org/abs/1909.04812.
Justin W. Hart is an assistant professor of practice in the Department of Com- puter Science at The University of Texas at Austin and the assistant director of the UT Austin Robotics Consortium.
Richard G. Freedman is a researcher at Smart Information Flow Technologies (SIFT), LLC and a Ph.D. candidate in the College of Information and Computer Sciences, University of Massachusetts Amherst.
Ross Mead is the Founder and CEO of Semio AI, Inc.
Nick DePalma is a visiting research scientist at Facebook Artificial Intelligence Research (FAIR).
Emmanuel Senft is a Research Associate at the University of Wisconsin-Madison in the Human-Computer Interaction Lab.
Elin A. Topp is a Reader (Associate Professor) in the group for Robotics and Semantic Systems (RSS), Dept of Computer Science, at Lund University, Swe- den.
Luca Iocchi is Full Professor at Dept. of Computer, Control and Management Engineering of Sapienza University of Rome, Italy.
Matteo Leonetti is a Lecturer (Assistant Professor) in the School of Computing at the University of Leeds, UK.