Reproduce “WANDR: Intention-guided Human Motion Generation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Markos Diomataris, Nikos Athanasiou, Omid Taheri, Xi Wang, Otmar Hilliges, Michael J. Black
Researchers from the Max Planck Institute for Intelligent Systems and ETH Zürich developed WANDR, a data-driven framework that generates natural human motions for 3D avatars to reach arbitrary 3D goals. WANDR achieves a 32% success rate in generalization tasks with 16% foot skating, outperforming prior data-driven methods by leveraging novel intention features and a unified training approach on diverse datasets.
Key insights
- Combining large-scale general locomotion data with smaller, precise goal-reaching data within a single framework significantly improves both the naturalness and generalization capabilities for goal-directed motion.
- Dynamically computed intention features are essential for guiding an avatar to arbitrary goals in a zero-shot manner, enabling robust performance without requiring explicit goal-reaching loss functions during training.
- Perceptual studies confirm that the motions generated by WANDR are perceptually close to real human movements, indicating high fidelity and realism in the synthesized actions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.