Section: New Results

Prediction of human error in robotics

The INRS provides a database of accidents at work, from which we can extract those concerning robotics. Many accidents are due to a deterioration of situational awareness. However, as there are many different causes and human factors are not well understood, it is very difficult for experts to provide probabilistic risk assessments. We proposed to simplify the problem by classifying the accidents according to the main demons that degrade the consciousness of the situation (Endsley model) and to use a Bayesian approach with the Noisy-Or nodes. We had already tried such an interpretation in the field of aeronautics. We propose to extend it to the field of robotics. Even if the approach remains empirical and approximate, it is possible to infer general probabilities of risk of human error leading to accidents and to deduce actions to reduce risks.

Related publications: [18]