Robots today are ill-equipped to predict how well they can perform a task or sense if a task is going as planned, let alone whether they did a good job once the task has been completed.
Humans, on the other hand, have the ability to judge how well they did something — throwing a baseball, turning a valve — or how well something has been built. With robots becoming more of a part of our daily lives, they will soon need this ability.
Researchers at Carnegie Mellon University, Brigham Young University, Tufts University and the University of Massachusetts have joined together in a five-year, $7.5 million project to develop methods and metrics for robotic self-assessment.
This self-assessment could be as simple as a robot being able to detect if a task was completed or it could be as complicated to include prediction and evaluation of proficiency. In other cases, it could include a robot providing an explanation to a human about its performance.
"You'd like the robot to be able to explain why it can or why it can't do a task," said Aaron Steinfeld, associate research professor in Carnegie Mellon University’s Robotics Institute. This may include a self-driving car telling why it can’t drop a passenger off at their required destination.
Part of the project will test how a robot approaches using dexterous search tasks for robots such as maneuvering limbs to investigate obscured items, manipulating objects to reveal contents and adversarial manipulation.
These tasks can be scaled up to applications such as robots deployed for emergency repairs, micro-drone swarms sent to map buildings and urban search and rescue.
