To act effectively in its environment, a cognitiverobot needs to understand the causal dependencies of allintermediate actions leading up to its goal. For example, thesystem has to infer that it is instrumental to open a cupboarddoor before trying to grasp an object inside the cupboard. In this paper, we introduce a novel learning method for extractinginstrumental dependencies by following the scientific cycleof observations, generation of causal hypotheses and testingthrough experiments.
Here is the video to the paper: https://youtu.be/eVThEsepfbw
Constantin Uhde, Nicolas Berberich, Karinne Ramirez-Amaro, and Gordon Cheng: The Robot as Scientist: Using Mental Simulation to Test Causal Hypotheses Extracted from Human Activities in Virtual Reality. IROS, 2020 (Finalist Best paper on Cognitive Robotics).