eduzhai > Applied Sciences > Engineering >

Identifying Causal Structure in Dynamical Systems

  • Save

... pages left unread,continue reading

Document pages: 14 pages

Abstract: We present a method for automatically identifying the causal structure of adynamical control system. Through a suitable experiment design and subsequentcausal analysis, the method reveals, which state and input variables of thesystem have a causal influence on each other. The experiment design builds onthe concept of controllability, which provides a systematic way to computeinput trajectories that steer the system to specific regions in its statespace. For the causal analysis, we leverage powerful techniques from causalinference and extend them to control systems. Further, we derive conditionsthat guarantee discovery of the true causal structure of the system and showthat the obtained knowledge of the causal structure reduces the complexity ofmodel learning and yields improved generalization capabilities. Experiments ona robot arm demonstrate reliable causal identification from real-world data andextrapolation to regions outside the training domain.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×