{
  "title": "TopoRetarget: set poses, optimize and track",
  "caption": "The author made this diagram from the source figure. Retargeting uses bone directions to set initial robot poses. Fixed shared-mesh relations constrain reference optimization. Policy tracking is a different subsequent stage. The example mesh keeps the same geometry. The diagram does not show optimization iterations, measured contacts or robot-result footage.",
  "alt": "Retargeting uses bone directions for initial robot poses, optimizes relations in a hand–object mesh and supplies references for a different tracking policy.",
  "sources": [
    "https://arxiv.org/abs/2606.16272",
    "https://toporetarget2026.github.io/TopoRetarget/"
  ],
  "limitations": [
    "The playback is an example. Playback time is not measured runtime.",
    "Graph vertices and edges are example geometry, not a sampled experimental mesh.",
    "The stage playback does not show the number of optimizer iterations or tracking success."
  ]
}