The International Journal of Robotics Research Study, Ahead of Publish.
We provide an extensive structure for examining and leveraging morphological balances in robot systems. These are inherent buildings of the robotic’s morphology, often observed in animal biology and robotics, which originate from the duplication of kinematic frameworks and the balanced circulation of mass. We show exactly how these balances include the robotic’s state area and both proprioceptive and exteroceptive sensing unit dimensions, causing the equivariance of the robotic’s formulas of movement and ideal control plans. Therefore, we identify morphological balances as an appropriate and formerly undiscovered physics-informed geometric prior, with considerable effects for both data-driven and logical techniques utilized in modeling, control, evaluation and layout in robotics. For data-driven techniques, we show that morphological balances can boost the example performance and generalization of artificial intelligence versions via information enhancement, or by using equivariant/invariant restrictions on the version’s style. In the context of logical techniques, we utilize abstract harmonic evaluation to disintegrate the robotic’s characteristics right into a superposition of lower-dimensional, independent characteristics. We validate our insurance claims with both artificial and real-world experiments carried out on bipedal and quadrupedal robotics. Finally, we present the repository MorphoSymm to assist in the useful use the concept and applications described in this job.
发布者：Daniel Ordonez Apraez，转转请注明出处：https://robotalks.cn/morphological-symmetries-in-robotics/
