Hand-Object Interaction & Affordance
Benchmarking Hand-Object Interaction Models Across Diverse Hand Morphologies and Embodiments
Generated automatically from the limitations stated in 7 papers (CVPR, NeurIPS, ECCV), listed under Evidence. It is not a paper, and it does not come from papers submitted to CSPaper.
The problem
Existing hand representation and hand-object interaction models are predominantly evaluated on small demographic cohorts or fixed kinematic setups (e.g., standard anthropomorphic models like MANO or fixed robotic hands like ShadowHand). Because existing pipelines implicitly bind kinematics, palm-to-finger ratios, or subject identities into their representations, their performance on unseen human hand proportions and non-standard robotic end-effectors remains untested. As a result, it is unknown whether current interaction and affordance estimators degrade gracefully or fail catastrophically when applied outside their narrow training morphologies.
Why it matters
Establishes the first rigorous empirical baseline for how hand pose, mesh reconstruction, and affordance estimators generalize across diverse human demographics and robotic hand morphologies, providing clear evaluation standards for cross-embodiment dexterity.
Ways to approach it
Prior-work checks are free with an account. Results someone already ran are shown to everyone.
- 1
Conduct a cross-dataset and cross-demographic evaluation of leading 3D hand and affordance models across diverse anthropometric benchmarks (varying palm-to-finger ratios, hand sizes, and ethnic/age diversity), measuring 3D joint error, mesh surface distance, and contact affordance transfer.
- 2
Build a standardized cross-embodiment benchmark spanning simulated robotic hands (e.g., 3-finger, 4-finger, and non-anthropomorphic grippers) and human hands, measuring zero-shot and few-shot grasp/interaction transfer success rates.
- 3
Systematically evaluate morphology-conditioned fine-tuning versus modular kinematic decoupling across current VAE and regression baselines to quantify the minimum adaptation data required for unseen hand structures.
Have a different approach?
Describe how you would tackle this problem and we'll look for papers that already do it. Free; your text stays private.
Why it might fail
The study would fail if existing single-embodiment architectures are fundamentally too rigid to adapt without full architectural redesigns, turning an empirical evaluation into disparate engineering patches without clear comparative takeaways.
Evidence
Each paper's own statement of the limitation, verbatim.
- Grasp2Grasp: Vision-Based Dexterous Grasp Translation via Schrödinger BridgesNeurIPS 2025
Does not generalize to unseen hand morphologies at inference: the VAE decoder implicitly encodes the target hand's kinematics, requiring retraining or fine-tuning for a new hand
- Contact Map Transfer with Conditional Diffusion Model for Generalizable Dexterous Grasp GenerationNeurIPS 2025
Validated only on ShadowHand (plus human hand); generalization to cross-embodiment hands with different finger counts and morphologies is untested and left to future work
- Dexterous Grasp TransformerCVPR 2024
Evaluated only in simulation (DexGraspNet, Isaac Gym) on ShadowHand; real-robot validation and non-fixed hand morphologies are not demonstrated
- 3D Hand Sequence Recovery from Real Blurry Images and Event StreamECCV 2024
EBH was captured from only 10 individuals, so shape diversity is limited and generalization to other hand shapes is uncertain
- Handy: Towards a High Fidelity 3D Hand Shape and Appearance ModelCVPR 2023
Dataset is demographically skewed (82% White, 2% Black, 9% Asian, 7% Mixed), so 'universal hand model' claim is only partially supported
- Affordance Diffusion: Synthesizing Hand-Object InteractionsCVPR 2023
Layout parameterization fixes the palm-size to forearm-width ratio to a constant mean over the training set, limiting shape variation
- DeepHandMesh: A Weakly-supervised Deep Encoder-Decoder Framework for High-fidelity Hand Mesh ModelingECCV 2020
Comparison with MANO on public 3D hand datasets was not possible due to the personalized assumption; quantitative evaluation is only on their own 4-subject dataset
Nearest existing work
- AffordPose: A Large-Scale Dataset of Hand-Object Interactions with Affordance-Driven Hand PoseICCV 2023
- Measuring Generalisation to Unseen Viewpoints, Articulations, Shapes and Objects for 3D Hand Pose Estimation under Hand-Object InteractionECCV 2020
- Dense Hand-Object(HO) GraspNet with Full Grasping Taxonomy and DynamicsECCV 2024
- Learning Joint Reconstruction of Hands and Manipulated ObjectsCVPR 2019
- GanHand: Predicting Human Grasp Affordances in Multi-Object ScenesCVPR 2020
- Authentic Hand Avatar from a Phone Scan via Universal Hand ModelCVPR 2024
- HOLD: Category-agnostic 3D Reconstruction of Interacting Hands and Objects from VideoCVPR 2024
- HTML: A Parametric Hand Texture Model for 3D Hand Reconstruction and PersonalizationECCV 2020
- Handy: Towards a High Fidelity 3D Hand Shape and Appearance ModelCVPR 2023
- Cross-Embodied Co-Design for Dexterous HandsICLR 2026
- MANUS: Markerless Grasp Capture using Articulated 3D GaussiansCVPR 2024
- DeepHandMesh: A Weakly-supervised Deep Encoder-Decoder Framework for High-fidelity Hand Mesh ModelingECCV 2020
- ContactPose: A Dataset of Grasps with Object Contact and Hand PoseECCV 2020
- Overcoming the Trade-Off Between Accuracy and Plausibility in 3D Hand Shape ReconstructionCVPR 2023
- FreiHAND: A Dataset for Markerless Capture of Hand Pose and Shape From Single RGB ImagesICCV 2019
Related open problems
Hand-Object Interaction & Affordance
Robust Hand-Object Interaction Learning Beyond Cascaded Parametric Pose Estimators
Downstream hand-object interaction (HOI) and affordance models are universally bottlenecked by a strict reliance on upstream off-the-shelf 3D hand pose estimators (e.g., FrankMocap, HaMeR, MANO parameterizations) and bounding-box detectors. When these upstream tools encounter severe object occlusions, fast manipulation dynamics, or egocentric perspective distortion, their estimation errors cascade directly into downstream HOI training, causing catastrophic mis-projections and synthetic contact artifacts. Because downstream methods require clean parametric 3D meshes as preconditions, massive repositories of in-the-wild manipulation videos cannot be utilized without fragile filtering heuristics that discard over half the data.
Hand-Object Interaction & Affordance
Robust 3D Hand-Object Pose and Interaction Estimation Under Upstream 2D Segmentation and Localization Noise
Current hand-object interaction and 3D hand pose estimation models rely heavily on upstream 2D bounding box crops, hand segmentation masks, or pre-extracted interaction regions. In real-world egocentric scenes, upstream 2D segmenters and detectors degrade severely due to object occlusions, motion blur, and cluttered backgrounds, causing downstream 3D estimation to fail or hallucinate. Because models are trained and benchmarked assuming near-ideal 2D crops or masks, existing pipelines cannot operate reliably in autonomous, in-the-wild video streams.
Hand-Object Interaction & Affordance
Fine-Grained 3D Hand Affordance Grounding and Generation for Small-Part Articulated Interactions
Current affordance learning pipelines rely on self-training loops, video extraction, or upstream generative HOI models that fail when contact targets are small functional components such as bottle caps, zippers, or rotary knobs. Because generated hand poses easily drift to adjacent object surfaces, models exhibit catastrophic failure on precision affordances (e.g., 0% accuracy on pull or sub-50% on twist), propagating noisy pseudo-labels across unobserved geometries. Downstream manipulation systems are consequently blocked from executing precision tasks like uncapping, pulling sliders, or fine bimanual manipulation.
Hand-Object Interaction & Affordance
Hand-Object Affordance Estimation Beyond Clean Pre-Scanned Scenes and Predefined 3D Grids
Current methods for 3D hand-object interaction and affordance estimation depend strictly on clean geometric preconditions: high-fidelity 3D scans, pre-detected object instances, static overhead views, or tightly predefined 3D bounding grids. In practical manipulation and augmented reality scenarios, environments are unsegmented, viewpoints are dynamic, and point clouds are noisy and incomplete. Relying on manually tuned grid volumes and pre-extracted meshes prevents existing models from operating directly on raw, uncurated sensor streams.