SynManDex: Synthesizing Human-like Dexterous Grasps from Synthetic Human Pre-Grasps

Noosaga Placements

• Robot Manipulationsubfield95%
  The paper focuses on generating dexterous grasps and manipulation trajectories for robotic hands, which is the core of robot manipulation.
  SynManDex, a synthetic pipeline that uses generated human pre-grasps as affordance-aware proposals and resolves the final contacts with robot-native optimization.The resulting keyframes support both grasp-and-lift demonstrations and various prehensile manipulation tasks such as tea pouring, photo taking, and flute playing.
• Learning-Based Manipulationframework90%
  The paper uses a learning-based approach to generate grasps and train a policy, which falls under the Learning-Based Manipulation framework.
  SynManDex, a synthetic pipeline that uses generated human pre-grasps as affordance-aware proposals and resolves the final contacts with robot-native optimization.Policies trained only on the generated synthetic demonstrations achieve 80.7% simulated zero-shot success.
• Robot Learningsubfield90%
  The paper uses generated demonstrations to train a closed-loop point-cloud policy for dexterous manipulation, which is a robot learning task.
  Policies trained only on the generated synthetic demonstrations achieve 80.7% simulated zero-shot success and 25/30 successes on a tabletop manipulation real-world benchmark.Admitted rollouts train a closed-loop point-cloud policy.
• Learning from Demonstrationframework85%
  The paper uses generated demonstrations to train a policy via imitation learning, which is a form of Learning from Demonstration.
  Admitted rollouts train a closed-loop point-cloud policy.Policies trained only on the generated synthetic demonstrations achieve 80.7% simulated zero-shot success.
• Roboticssubfield80%
  The paper bridges human hand-object interaction priors with robotic grasp synthesis, a core AI-robotics problem.
  Human hand-object interactions encode functional intent, but direct transfer to robotic hands often fails under morphology, contact, and reachability constraints.We present SynManDex, a synthetic pipeline that uses generated human pre-grasps as affordance-aware proposals.
• Imitation Learningframework80%
  The policy is trained on demonstration data, which is a form of Imitation Learning.
  Admitted rollouts train a closed-loop point-cloud policy.Policies trained only on the generated synthetic demonstrations achieve 80.7% simulated zero-shot success.
• Diffusion Modelsframework70%
  The paper uses a diffusion model to generate human pre-grasps, which is a Diffusion Models framework.
  SynManDex-Human is an object-conditioned diffusion model trained on hand-object resources such as GRAB and ContactPose.
• Computer Visionsubfield50%
  The paper uses a diffusion model conditioned on object meshes to generate human pre-grasps, which involves computer vision techniques, but the focus is on robotics.
  SynManDex-Human is an object-conditioned diffusion model trained on hand-object resources such as GRAB and ContactPose.

Abstract

Human hand-object interactions encode functional intent, but direct transfer to robotic hands often fails under morphology, contact, and reachability constraints. We present SynManDex, a synthetic pipeline that uses generated human pre-grasps as affordance-aware proposals and resolves the final contacts with robot-native optimization. SynManDex samples object-conditioned digital human pre-grasps, retargets them to dexterous robotic hand poses, optimizes force-closure contacts on the target embodiment, and admits trajectories that pass checks from each step. The resulting keyframes support both grasp-and-lift demonstrations and various prehensile manipulation tasks such as tea pouring, photo taking, and flute playing, designed via VLM agents. As a result, SynManDex combines high grasp quality (86.4\% grasp stability) with 4.67/5 human-likeness (93.4\%). It achieves 80.7\% successes in simulation and 25/30 (83.3\%) real-robot successes when applied to a 36-DOF bimanual dexterous robotic platform.

Paper Context