Protection from Human Error

By Kevin S. Pratt and Robin R. Murphy

NOTE: This is an overview of the entire article, which appeared in the December 2012 issue of the IEEE Robotics & Automation Magazine.
Click here to read the entire article.

The use of robotics in the biomedical field is rapidly increasing. While this survey article covers a wide range of applications of mobile robots and manipulator arms, its applicability to the life sciences is straightforward. The article surveys and categorizes aspects of 32 systems using a new taxonomy including control elements and interface characteristics. The specific control focus in this article is “guarded motion”, which the authors define as “a method for monitoring and addressing safety constraints for human-directed operations when the human has inferior knowledge (compared to the robot) of the robot’s pose and relation to its environment. In essence, guarded motion is a type of human-in-the-loop control where the robot guards itself from unintended consequences of human directives (collision avoidance, unstable configurations, excessive force, unnecessary power consumption, etc.).”

The article first reviews the history of guarded motion, which is found in research from the 1970’s onward in robotic arms and space program rovers. This background is evident in many of the 32 systems they include in the study, although there are a fair number of biomedical applications, such as wheelchairs.

The bulk of the article presents the taxonomy of the classified systems, based on 5 axes:

• Autonomy Intervention Criteria: details when the system intervenes with the commands issued by the operator.
• Command Integration Method: covers how the system integrates its commands with the user’s commands.
• Monitored Condition: describes which variables and environmental conditions the system monitors and accounts for.
• Interface Modality: accounts for how the system presents information to the operator.
• Display Preprocessing: discusses how much processing is performed on sensor data before it is presented to the operator.

Each of these axes is explored in the article, which provides insight into the range of options in that segment of control or display, and discusses the limitations and advantages of the strategies that have been used historically. Some particular attention is directed to biomedical applications, for instance in a discussion of wheelchair control methodologies.

Following the taxonomic analysis, the article makes some general observations on guarded motion controls in the analyzed systems. A number of open research questions are raised, for instance considering situations where guarded motion may not work well.

The article concludes with some general findings that “can help inform the incorporation of guarded motion into a robot system.” An extensive set of references is also provided.

ABOUT THE AUTHORS

Kevin S. Pratt (kpratt@gmail.com) is with Texas A&M, College Station, Texas, USA.

Robin R. Murphy (murphy@cse.tamu.edu) is is with Texas A&M, College Station, Texas, USA.