Can Robots and Humans Get Along?
June / July 2007
Volume 5 Number 3
Now that robots have moved into the mainstream as vacuum cleaners, lawn mowers, autonomous vehicles, tour guides and even pets, it is important to consider how everyday people interact with them—€”today and in the future—€”as robots become increasingly useful and sophisticated.
A robot is really just a computer, but researchers are beginning to understand that human-robot interactions are much different than human-computer interactions. While the metrics used to evaluate the human-computer interaction (usability of the software interface in terms of time, accuracy and user satisfaction) may also be appropriate for human-robot interactions, we need to determine if additional metrics should be considered.
Important differences between human-computer and human-robot interactions include the role of the user and the mobility and autonomy of the robot. Humans interact with robots in many roles. One person might operate robots. Another might supervise robots or teams of humans and robots, while still another might be a member of the robot-human team. Programmers and mechanics are needed to fix the robots and test the repairs. Each of the roles involves different types of interactions and therefore different user interfaces.
Then there is the bystander, who has no direct role with the robot but who interacts with robots because they occupy the same space at the same time. For instance, drivers will need to know how autonomous robotic vehicles will behave on the road. Do the same rules apply to robots as apply to human drivers? Or is it necessary to define some new rules?
Robot operations vary from complete tele-operation to total autonomy. The level of autonomy can be changed based on the task the robot is doing, the environment in which the task is being performed and the operator's confidence in the robot's abilities. Regardless of role, the user needs to be aware of the level of autonomy under which the robot is operating. The interface may provide the capability for users and robots to negotiate who should have control at a given time. It may be that a robot stuck in a tight place is better suited to use its sensors to escape because it could be difficult for the operator to have enough situational awareness to "rescue" the robot. The user of robots operating in an autonomous manner must be kept informed so she can intervene quickly if necessary.
Another difference between human-robot and human-computer interactions is that computers by and large are stationary and co-located with the user. Robots have the advantage of being able to move around in the world, often in places that are too dangerous for humans. If an operator is located remotely, she will need to know where her robots are, what they are doing and the problems they may encounter. While sensors can certainly provide some of this information, the robot user must have certain knowledge so that she can act quickly to avoid problems.
It is difficult and expensive to perform user studies of human-robot interfaces. First, the robot platforms and the user interface software must be fairly robust. If we are comparing two interfaces, both need to be developed to the same level of maturity, and the skill and training of the users should be comparable. Enough space for conducting the study has to be available. For larger robots, safety is a consideration for people and robots. And finally, it is extremely difficult to exactly repeat an experiment. A simple task of navigating between two points rarely results in robots taking the same path.
Some of the ways to study human-robot interaction include user studies, field observations, competitions and developing baselines for comparison. User studies and field observations are similar to those types of studies in human-computer interaction. Examples of competitions and baseline comparison follow.
To obtain an overview of user interfaces for robotic platforms and to evaluate designs that worked, researchers collected data for several years at an Urban Search and Rescue competition developed by the National Institute of Standards and Technology. This competition is held yearly in conjunction with Robocup and is conducted in an arena designed to resemble a collapsed building. Many victims are hidden in the arena. Victims can be recognized by sight (manikins) or by sensors (heat, sound and motion).
With the permission of the competing teams, the competition was videotaped to provide a record of what really happened. In addition, tapes of the user interface revealed what the operator was seeing and doing. Researchers analyzed the data for "critical incidents," times when the robot harmed the environment, the victim or itself. This analysis led to the establishment of design guidelines for information presentation and interactions in the operator interface.
NIST developed a baseline comparison technique to evaluate user interfaces for robots that dispose of explosive ordnance. In conjunction with subject matter experts, researchers devised small tasks that would be used in a typical explosive ordnance disposal (EOD) mission: climbing a set of stairs, picking up a suitcase, looking under a sofa for a suspicious device.
Expert users of different types of ordnance disposal robots completed the tasks as quickly as possible and measured the time and accuracy of each task. An organization purchasing an EOD robot can have several staff trained to use the robot and perform these standard tasks. The organization can assess the performance of its users compared with the experts and decide if this level of performance will be suitable for their mission.
In human-computer interaction, we normally think of the computer as subservient to the human. In human-robotic interaction, we are beginning to look at how teams of humans and robots could work together and how to measure the effectiveness of these teams.
A project at NASA's Ames Research Center developed an operating system that facilitates cooperation between astronauts and robots in an extra-vehicular activity environment. The operating system tracks who is doing what and who has what resources. Requests for help from both robots and humans are accepted by the operating system and given to the appropriate teammate.
A pilot study conducted during this project developed a measure that can be used to evaluate how a team of humans and robots work together. Human teams are able to do the majority of tasks faster than a team of robots. But a team of robots working alone eliminates humans from exposure to risky situations, unless a robot runs into trouble and a human must come to the rescue. Although most robots have limitations and need human intervention at some point, the study showed that teams of humans and robots can accomplish more in a given time than either humans or robots alone.
These examples involve robots performing work-related tasks. Other research assesses interfaces for more social robots, such as pets and companions, caregivers, and receptionists. A number of researchers in social robotics have looked at metrics such as the user's acceptance, engagement, and enjoyment.
The human-robot interaction community just had its second conference and announced the call for papers for the third. As more robots become a part of our daily lives, researchers will have more work to do. The most important focus will be on user interfaces for robot capabilities that are increasing every day.
Those interfaces must accommodate a wide range of users, tasks, and autonomy levels.
How well these robots are integrated into our society will depend on how well we can work and play with them.
Jean Scholtz is retired from the National Institute of Standards and Technology, where she conducted pioneering work in human-robot interaction. She joined Pacific Northwest National Laboratory in 2006 as an expert in information analytics.
Copyright © 2012 | Innovation America