Discovery Sagaサイレントキーワード俯瞰

graph learning / planning / networks / optimization / robotics / succinct representation / succint representation

The foundations for learning and optimization of graphs via enumerative queries and a their application benchmarks in computer-aided design and robotics were proposed: (1) the sampling and learning mechanism for obstacle-avoiding lattice paths the enumerative encoding and gradient-free heuristics, (2) the learning of adaptive locomotion gaits for six-legged robots under conditions of leg failure by using the enumerative queries, (3) the compact representation of convolution graphs by broadcast networks and its application to sound classification, (4) the representation of curvature in the folding of planar membranes, allowing the representation of folding structures by compact planar graphs, (5) the development of multi-modal path planning, allowing the efficient computation of collision-free paths for multi-agent systems in lattice paths, (6) the sampling of graph-based cable-driven robotic structures, enabling the design of cable-driven mechanical structures by enumerative sampling of the graph search space, (7) the study of fairness functionals for smooth path planning in mobile robots, allowing the efficient evaluation of smoothness over a network of multiple robotics trajectories, and the (8) applications of graph-based sampling and modeling of robotic structures, having the potential for enumerative and efficient queries of robotic topologies.

learning / networks / graph representation / optimization / network design / product design / path planning / evolutionary computing / learning networks / design / control / evolutionary computation / representations / combinatorial / pattern recognition / succinctness / computer-aided design / modularity / succint representation / robotics

本研究では簡潔な構造を持つ学習ネットワークとその応用を行った。一方では、最先端の効率を達成する有向ネットワーク、無向ネットワーク、モジュラーネットワークおよび可変サイズネットワークの簡潔な構造が可能になった。そして経路計画とCAD問題における応用は提案したアルゴリズムの優れた性能、効率とスケーラビリティを示すことを明らかにした。