Interactive Programming Framework for Learning-Enabled Robots

Advanced robots are scaling into diverse, non-expert markets. However, the industry lacks a scalable model to maintain and re-task or reprogram these machines to fit unique user environments. This highlights a technical bottleneck that is most acute for learning-enabled robots, which create a highly inefficient and unscalable business model. Because each robot learns uniquely from its specific user, standardized customer support is impossible. Minor environmental deviations from the original training data can cause catastrophic, expensive operational accidents or even failures.

 

Researchers at Arizona State University have developed an innovative user-friendly framework that enables non-experts to re-task learning-enabled robots safely and effectively. This programming framework allows users without coding expertise to an intuitive graphical interface and adaptive models to program robots based on particular robot capabilities. It leverages a query-response interface to gather robot performance data and dynamically updates probabilistic models of robot skills. These models guide motion planning and task execution, while the system offers clear feedback on errors and suggests training tasks to enhance user proficiency and robot effectiveness.

 

This user-friendly framework enables non-experts to program learning-enabled robots through an intuitive graphical interface and adaptive models.