In the direction of a Causal Probabilistic Framework for Prediction, Motion-Choice & Explanations for Robotic Block-Stacking Duties
Authors: Ricardo Cannizzaro, Jonathan Routley, Lars Kunze
Summary: Uncertainties in the true world imply that’s unattainable for system designers to anticipate and explicitly design for all situations {that a} robotic would possibly encounter. Thus, robots designed like this are fragile and fail exterior of highly-controlled environments. Causal fashions present a principled framework to encode formal information of the causal relationships that govern the robotic’s interplay with its surroundings, along with probabilistic representations of noise and uncertainty usually encountered by real-world robots. Mixed with causal inference, these fashions allow an autonomous agent to know, motive about, and clarify its surroundings. On this work, we concentrate on the issue of a robotic block-stacking job because of the elementary notion and manipulation capabilities it demonstrates, required by many purposes together with warehouse logistics and home human assist robotics. We suggest a novel causal probabilistic framework to embed a physics simulation functionality right into a structural causal mannequin to allow robots to understand and assess the present state of a block-stacking job, motive concerning the next-best motion from placement candidates, and generate post-hoc counterfactual explanations. We offer exemplar next-best motion choice outcomes and description deliberate experimentation in simulated and real-world robotic block-stacking duties.