A curated list of data-centric methods for Vision-Language-Action models and robot foundation models: collection, simulation, digital twins, cross-embodiment augmentation, neural trajectories, RL-generated rollouts, cleaning, preprocessing, annotation, and benchmarks.
|
|
|
|
📑 Full Table of Contents (click to expand)
| Tag | Meaning |
|---|---|
Data |
Dataset or generated trajectories are released. |
Code |
Collection, conversion, generation, training, or evaluation code is released. |
Method |
Pipeline is described, but artifacts may be private or unclear. |
Gated |
Access or license restrictions apply. |
Weights |
Model checkpoints are released. |
Benchmark |
Evaluation tasks, metrics, or benchmark data are released. |
single-arm · bimanual · mobile-manipulator · humanoid · dexterous-hand · cross-embodiment · robot-free · simulation · real-world · latent-action · world-model · RL-generated
| Paper | Year | Focus |
|---|---|---|
| Vision-Language-Action in Robotics: A Survey of Datasets, Benchmarks, and Data Engines | 2026.04 |
Data-centric VLA survey. |
| 3D Gaussian Splatting in Robotics: A Survey | 2024.10 |
Survey of 3DGS scene representations, rendering, and robotics applications. |
| Vision Language Action Models in Robotic Manipulation: A Systematic Review | 2025.07 |
Systematic review of VLA models, datasets, and simulation platforms for manipulation. |
| Vision-Language-Action Models for Robotics: A Survey | 2025 |
Broad VLA survey. |
| Survey of Vision-Language-Action Models for Embodied Manipulation | 2025.08 |
VLA methods and embodied manipulation. |
| What Matters in Building Vision-Language-Action Models for Generalist Robots | 2025 |
Empirical study on VLA data and design choices. |
| A Tutorial Note on Collecting Simulated Data for Vision-Language-Action Models | 2025.08 |
Simulated VLA data collection. |
| Repository | Focus |
|---|---|
| Open X-Embodiment | Multi-embodiment robot data. |
| VLA Datasets, Benchmarks, and Data Engines | Data-centric VLA datasets, benchmarks, and data-engine taxonomy. |
| UMI Robot Dataset Community | UMI-style datasets. |
| Resource | Focus |
|---|---|
| LeRobot Documentation | Robot datasets, policies, and tooling. |
| OpenVLA | OXE preprocessing and VLA fine-tuning. |
| VLA Foundry | Unified LLM/VLM/VLA training and robot data preprocessing. |
| Work | Year | Embodiment | Artifact | Notes |
|---|---|---|---|---|
| Open X-Embodiment / RT-X | 2023.10 |
cross-embodiment | Data |
Multi-robot dataset for cross-embodiment learning. |
| BridgeData V2 | 2023.08 |
single-arm | Data |
Real-world WidowX manipulation data. |
| DROID | 2024.03 |
single-arm | Data Code |
In-the-wild robot manipulation demonstrations. |
| LeRobot | 2026.02 |
cross-embodiment | Data Code |
Open robot data and policy ecosystem. |
| RoboMIND | 2025 |
cross-embodiment | Data |
Real-world manipulation trajectories across robots and tasks. |
| RoboMIND 2.0 | 2025.12 |
bimanual, mobile-manipulator, cross-embodiment | Data Benchmark |
310K+ real dual-arm trajectories across six embodiments, plus tactile/mobile manipulation and simulated trajectories. |
| AgiBot World | 2025 |
mobile, humanoid, bimanual | Data |
Large-scale real robot corpus. |
| AgiBot World Colosseo | 2025.03 |
bimanual, dexterous, cross-embodiment | Data Code Weights |
Large-scale manipulation platform with 1M+ trajectories and GO-1 policy artifacts. |
| Galaxea Open-World Dataset | 2025 |
mobile dual-arm | Gated Data |
Open-world behavior data. |
| Action100M | 2026.01 |
human / internet video | Data |
O(100M) dense open-vocabulary action annotations from instructional videos for world-model and embodied pretraining. |
| OpenEAI Dataset | 2026 |
cross-embodiment | Data |
Unified VLA pretraining corpus aggregating Open X-Embodiment, UMI Community, and related embodied data sources. |
| RH20T | 2023 |
single-arm | Data |
Robot-human multimodal manipulation. |
| CLAMP | 2025.05 |
tactile, robot-free | Data Code |
In-the-wild multimodal haptic dataset collected with a low-cost open-source device. |
| Kaiwu | 2025.03 |
human, robot, multimodal | Data |
Synchronized multimodal manipulation data with pressure, audio, mocap, gaze, EMG, and fine-grained labels. |
| RoboSet / RoboHive | 2023 |
arms, hands | Data Code |
Real and simulated robot learning data. |
| RoboNet | 2020 |
multi-robot | Data Code |
Shared multi-robot experience dataset. |
| Work | Year | Embodiment | Artifact | Notes |
|---|---|---|---|---|
| Humanoid Everyday | 2025 |
humanoid | Data |
Everyday humanoid manipulation data. |
| Unitree UnifoLM-WBT Dataset | 2025 |
humanoid | Data |
Whole-body teleoperation data. |
| Fourier ActionNet | 2026 |
humanoid, bimanual | Data |
Dexterous bimanual humanoid dataset. |
| DexCap | 2024.03 |
dexterous-hand | Code Method |
Portable hand mocap and retargeting. |
| DexUMI | 2025.05 |
dexterous-hand | Data Method |
Robot-free dexterous collection. |
| XL-VLA | 2026 |
dexterous-hand | Data Code |
Cross-hand teleoperation data and latent action representation. |
| RealSource-World | 2025 |
humanoid, arms | Data |
Real robot data release. |
| Benchmark | Year | Setting | Artifact | Notes |
|---|---|---|---|---|
| LIBERO | 2023 |
single-arm, simulation | Data Code Benchmark |
Lifelong language-conditioned manipulation benchmark. |
| RLBench | 2019 |
single-arm, simulation | Data Code Benchmark |
Large-scale vision-guided manipulation suite built on CoppeliaSim/PyRep. |
| CALVIN | 2021 |
single-arm, simulation | Data Code Benchmark |
Long-horizon language-conditioned manipulation benchmark. |
| SimplerEnv | 2024 |
single-arm, simulation | Code Benchmark |
Sim-real evaluation framework for Google Robot and WidowX-style policies. |
| ManiSkill2 | 2023 |
arms, mobile, dexterous, simulation | Data Code Benchmark |
Unified benchmark for generalizable manipulation skills. |
| RoboCasa | 2024 |
household manipulation, simulation | Data Code Benchmark |
Photorealistic household manipulation benchmark built on robosuite. |
| RoboTwin 2.0 | 2025 |
bimanual, simulation | Data Code Benchmark |
Dual-arm benchmark and scalable data generator with domain randomization. |
| VLABench | 2025 |
long-horizon manipulation, simulation | Data Code Benchmark |
Language-conditioned manipulation benchmark for long-horizon reasoning. |
| Kinetix | 2024 |
2D physics, RL | Code Benchmark |
Open-ended physics-control benchmark; adjacent to robot policy evaluation. |
| RoboTwin 1.0 | 2024 |
bimanual, simulation | Data Code Benchmark |
Early RoboTwin dual-arm benchmark with generative digital twins. |
| LIBERO-Plus | 2026 |
single-arm, robustness | Data Benchmark |
Robustness extension of LIBERO. |
| LIBERO-Pro | 2025 |
single-arm, robustness | Code Benchmark |
LIBERO extension for evaluation beyond memorization. |
| RoboArena | 2025 |
real-world, distributed | Benchmark |
Community-run real-world benchmark for generalist robot policies. |
| MIKASA-Robo | 2025 |
tabletop manipulation, memory | Code Benchmark |
Memory-intensive tabletop manipulation benchmark. |
| RoboCerebra | 2025 |
long-horizon manipulation | Data Benchmark |
Long-horizon benchmark for planning, reflection, and memory. |
| LIBERO-Mem | 2026 |
single-arm, memory | Benchmark |
Non-Markovian object-centric memory benchmark. |
| RoboChallenge | 2025 |
real-world manipulation | Benchmark |
Real-robot evaluation benchmark for embodied policies. |
| RoboMME | 2026 |
memory-augmented manipulation | Data Code Benchmark |
Benchmark for memory-augmented robotic manipulation. |
| Work | Year | Embodiment | Artifact | Core Mechanism |
|---|---|---|---|---|
| ALOHA | 2023 |
bimanual | Code Data |
Low-cost bimanual teleoperation. |
| Mobile ALOHA | 2024.01 |
mobile bimanual | Code Data |
Mobile bimanual teleoperation. |
| ALOHA 2 | 2024 |
bimanual | Code |
Improved low-cost bimanual teleop hardware. |
| UMI | 2024.02 |
robot-free, single-arm | Code Data |
Handheld gripper for robot-free demonstrations. |
| UMI Data Community | 2025 |
robot-free | Data |
UMI-style dataset hub. |
| FastUMI | 2024.09 |
robot-free | Data Code |
Scalable UMI-style collection. |
| UMI-3D | 2026.04 |
robot-free | Method |
UMI with 3D spatial perception. |
| DexUMI | 2025.05 |
dexterous-hand | Data Method |
Human hand collection with robot-hand conversion. |
| DexCap | 2024.03 |
dexterous-hand | Code Method |
Hand mocap and retargeting. |
| DEX-Mouse | 2026.04 |
dexterous-hand | Method |
Low-cost force-feedback dexterous teleop. |
| SABER | 2025 |
ego, hands, humanoid | Data Method |
Human behavior capture to embodied action supervision. |
| Lucid-XR | 2026.04 |
dexterous, robot-free, simulation | Method |
XR-headset physics simulation, human-to-robot retargeting, and physics-guided video generation for synthetic manipulation data. |
| GELLO | 2023.09 |
single-arm, bimanual | Code Method |
Low-cost kinematically matched teleoperation devices for high-quality demonstrations. |
| RoboWheel | 2025.12 |
cross-embodiment | Method |
Human-object interaction reconstruction for robot learning. |
| MV-UMI | 2026 |
cross-embodiment | Method |
Multi-view UMI-style interface for cross-embodiment learning. |
| Touch in the Wild | 2025.07 |
robot-free, tactile | Method |
Portable visuo-tactile gripper for fine-grained in-the-wild demonstrations. |
| Work | Year | Embodiment | Artifact | Core Mechanism |
|---|---|---|---|---|
| MimicGen | 2023.10 |
single-arm | Data Code |
Object-centric trajectory retargeting. |
| SkillMimicGen / SkillGen | 2024.10 |
single-arm | Method |
Skill segmentation and motion-planner stitching. |
| DexMimicGen | 2024.10 |
bimanual, dexterous | Data Code |
Bimanual dexterous demo generation. |
| SoftMimicGen | 2026.03 |
deformable, bimanual, humanoid | Method |
MimicGen-style automated data generation for rope, towel, tissue, stuffed-animal, surgical, and humanoid deformable manipulation tasks. |
| DiffGen | 2024.05 |
simulation, manipulation | Method |
Differentiable physics, differentiable rendering, and VLM objectives for automatic robot demonstration generation. |
| IntervenGen | 2024.05 |
simulation, real-world | Method |
Expands a small number of human corrective interventions into synthetic recovery demonstrations for robust imitation learning. |
| CyberDemo | 2024.02 |
dexterous-hand, simulation | Data Method |
Simulated human demonstrations with domain and task augmentation for real-world dexterous manipulation transfer. |
| RoboGen | 2023.11 |
manipulation, locomotion | Code Method |
Generative simulation for tasks and supervision. |
| GenSim2 | 2024.10 |
simulation, articulated objects | Code Method |
Multimodal LLM task generation with planning and RL solvers that generate demonstrations for articulated manipulation. |
| Scaling Up and Distilling Down | 2023.07 |
simulation, language-conditioned | Code Data |
LLM-guided planners generate diverse language-labeled robot trajectories that are distilled into a multitask visuomotor policy. |
| RoboCasa | 2024.06 |
household manipulation | Data Code |
Generated household tasks and demos. |
| RoboCasa365 | 2026.03 |
household, mobile manipulation | Data Method |
Large-scale household task suite. |
| RoboTwin | 2025.04 |
bimanual | Data Code |
Dual-arm digital-twin-style benchmark. |
| RoboTwin 2.0 | 2025.06 |
bimanual, cross-embodiment | Data Code |
Scalable dual-arm generator. |
| GenSim | 2023.10 |
simulation, manipulation | Code Method |
LLM-generated simulation tasks, curricula, and expert demonstrations. |
| Video2Policy | 2025.02 |
simulation, human / internet video | Method |
Reconstructs manipulation tasks from internet RGB videos and trains RL policies with LLM-generated rewards. |
| Learning Interactive Real-World Simulators | 2023.10 |
simulation, world-model | Method |
Learns interactive simulators from real-world data for zero-shot policy transfer. |
| Scaling Robot Learning with Semantically Imagined Experience | 2023.02 |
simulation, single-arm | Method |
Semantically imagined experience for scaling robot learning with generated variations. |
| MolmoB0T / MolmoBot-Engine | 2025 |
single-arm, mobile manipulator | Data Code |
Procedural simulation engine. |
| AGT-World | 2026.02 |
simulation, long-horizon | Method |
Affordance-graph task worlds with self-evolving VLM/geometric feedback for task and policy generation. |
| Generative Simulation for pHRI | 2026.04 |
physical HRI, simulation | Method |
Text2sim2real pipeline for synthesizing assistive pHRI scenes, soft-body humans, and robot motion trajectories. |
| CP-Gen | 2025.08 |
single-arm | Method |
Constraint-preserving keypoint generation. |
| DynaMimicGen | 2025.11 |
dynamic manipulation | Method |
Data generation for dynamic tasks. |
| MoMaGen | 2025.10 |
bimanual mobile | Method |
Reachability and visibility constrained generation. |
| HumanoidGen | 2025.07 |
humanoid, dexterous | Data Code |
LLM/MCTS-generated humanoid dexterous demos. |
| Work | Year | Embodiment | Artifact | Core Mechanism |
|---|---|---|---|---|
| DemoGen | 2025.02 |
arm, bimanual, dexterous | Code Method |
3D point-cloud scene and trajectory synthesis. |
| RoboSplat | 2025.04 |
3DGS, manipulation | Method |
Edits 3D Gaussians to generate one-shot visual demonstrations across object pose/type, camera, lighting, scene appearance, and embodiment changes. |
| RoboGSim | 2024.11 |
3DGS, simulation | Code Method |
Real2Sim2Real robotic simulator combining 3DGS reconstruction with physics-engine interaction for synthetic trajectories, scenes, objects, and viewpoints. |
| SplatSim | 2024.09 |
3DGS, simulation | Method |
Uses Gaussian splats as photoreal rendering primitives for zero-shot sim-to-real RGB manipulation policy training. |
| NeRF-Aug | 2025.09 |
NeRF, manipulation | Method |
Uses object-level neural radiance fields to synthesize photorealistic, 3D-consistent demonstrations for novel objects without collecting new human demos. |
| DISCOVERSE | 2025.07 |
3DGS, simulation | Code Method |
Open-source Real2Sim2Real simulator using 3DGS for high-fidelity rendering and MuJoCo for physics. |
| RoboSimGS | 2025.10 |
3DGS, simulation | Method |
Hybrid 3DGS-plus-mesh framework where 3DGS captures photoreal appearance and mesh primitives provide physically interactive manipulation assets. |
| A High-Fidelity Digital Twin for Robotic Manipulation Based on 3DGS | 2026.01 |
3DGS, digital twin | Method |
Converts sparse RGB 3DGS reconstructions into semantically labeled, collision-ready digital twins integrated with Unity, ROS2, and MoveIt. |
| EmbodiedGen / RoboSplatter | 2025.06 |
3DGS, embodied simulation | Code Method |
Generative 3D world engine with RoboSplatter for high-fidelity 3DGS rendering inside embodied simulation pipelines. |
| GaussGym | 2025.10 |
3DGS, locomotion | Code Data |
Real-to-sim framework that plugs 3DGS rendering into vectorized physics simulators for high-throughput pixel-based robot learning. |
| Real2Render2Real | 2025.05 |
robot arms | Method |
Real scan and human video to rendered robot demonstrations. |
| EMMA | 2025.09 |
real-world, single-arm | Method |
Generative visual transfer via multi-view consistent embodied manipulation video editing. |
| ERMV | 2025.07 |
multi-view, VLA | Method |
Edits 4D robotic multi-view image sequences with geometric, temporal, and robot-state consistency for VLA data augmentation. |
| AOMGen | 2025.12 |
articulated objects, simulation | Method |
Photoreal, physics-consistent demonstration generation for articulated objects from a single scan, demonstration, and digital assets. |
| Work | Year | Embodiment | Artifact | Core Mechanism |
|---|---|---|---|---|
| MIRAGE | 2024.02 |
cross-embodiment arms | Code Method |
Robot masking and visual inpainting. |
| RoVi-Aug | 2024.09 |
cross-embodiment arms | Code Method |
Robot appearance and viewpoint augmentation. |
| RoboEngine | 2025.03 |
cross-scene robot data | Method |
Robot segmentation and background generation. |
| H2R | 2025.05 |
human video to robot | Method |
Human hand keypoints to rendered robot motions. |
| From Generated Human Videos to Physically Plausible Robot Trajectories | 2025.12 |
humanoid | Method Benchmark |
Converts generated human videos into physically plausible humanoid trajectories through reconstruction, retargeting, and robust policy execution. |
| X-Humanoid | 2025.12 |
humanoid | Data Method |
Robotizes human videos into humanoid videos at scale using paired synthetic data and video-to-video generation. |
| OXE-AugE | 2025.12 |
cross-embodiment | Data Method |
Robot augmentation over Open X-Embodiment. |
| CEI: Cross-Embodiment Interface | 2026.01 |
arms, grippers, hands | Method |
Observation and action synthesis across embodiments. |
| R2RGen | 2025.10 |
real robot point clouds | Method |
Real-to-real 3D pointcloud/action augmentation. |
| XL-VLA | 2026 |
dexterous hands | Code Data |
Cross-hand latent action representation. |
| Being-H0 | 2025.07 |
human hands, dexterous | Method |
Human-video physical instruction tuning and motion tokenization. |
| Being-H0.5 | 2026.01 |
cross-embodiment | Method Weights |
Unified action space for embodiment transfer. |
| OmniRetarget | 2025.09 |
humanoid | Method |
Whole-body humanoid motion retargeting. |
| SPIDER | 2026 |
dexterous, humanoid | Method |
Physics-based retargeting for hands and humanoids. |
| Perceptive Humanoid Parkour | 2026.02 |
humanoid | Method |
Retargets and composes atomic human parkour skills with motion matching, then distills them into a depth-based real-robot policy. |
| Work | Year | Embodiment | Artifact | Core Mechanism |
|---|---|---|---|---|
| DreamGen | 2025.05 |
arms, humanoid | Method |
Video world model to action-labeled neural trajectories. |
| RLDX-1 | 2026.05 |
dexterous, humanoid | Code Weights Method |
Video generation, motion filtering, and inverse dynamics labels. |
| RoboCurate | 2026.02 |
humanoid, dexterous | Method |
Action-verified neural trajectory filtering. |
| EVA | 2026.03 |
bimanual, arms | Method |
Inverse-dynamics reward for executable videos. |
| TC-IDM | 2026.01 |
generated video, 3D motion | Method |
Tool-centric inverse dynamics model that converts world-model video plans into 6-DoF end-effector motions and controls. |
| NovaFlow | 2025.10 |
generated video, 3D flow | Method |
Distills generated videos into 3D actionable object flow and realizes it with grasp proposals, trajectory optimization, and particle dynamics. |
| TraceGen / TraceForge | 2025.11 |
cross-embodiment | Method |
3D trace-space world modeling and trace data pipeline. |
| Work | Year | Embodiment | Artifact | Core Mechanism |
|---|---|---|---|---|
| RLDG | 2024.12 |
real robot, single-arm | Code Method |
Trains task-specific RL specialists, rolls them out to generate high-quality data, and distills the data into generalist robot policies. |
| RDGen | 2026.05 |
sim-real, single-arm | Method |
Trains RL policies as structured trajectory generators, transfers them to real robots, and harvests successful rollouts as demonstrations for downstream VLA training. |
| PLD / Deployment-Aligned Data | 2026.05 |
sim-real, VLA post-training | Method |
Trains lightweight residual RL specialists, collects recovery trajectories from base-policy states, and distills them back into the VLA. |
| RL-Driven Data Generation for Dexterous Grasping | 2025.04 |
dexterous-hand, simulation | Method |
Uses a residual RL teacher skill to generate contact-rich grasping trajectories across object geometries for vision-action policy training. |
| Beyond Human Demonstrations | 2025.09 |
simulation, arms | Method |
Diffusion RL experts generate VLA training data. |
| Discover, Learn, and Reinforce | 2025.11 |
simulation, arms | Method |
Diverse RL rollouts for VLA pretraining. |
| OmniReset | 2025 |
arms, dexterous | Code Method |
Diverse resets, RL experts, and RGB distillation. |
| Work | Year | Artifact | Pipeline |
|---|---|---|---|
| Robust Learning from Demonstrations with Mixed Qualities Using Leveraged Gaussian Processes | 2019 |
autonomous driving, LfD | Method |
| ABot-M0 / UniACT | 2026.02 |
Method |
Cleans, standardizes, and balances heterogeneous public datasets into UniACT. |
| VLA Foundry | 2026.04 |
Code Method |
WebDataset sharding, normalization, action chunking, SE(3) actions, and multi-dataset stats. |
| HoloBrain-0 / RoboOrchard | 2026.02 |
Code Method |
Full-stack VLA infrastructure for data curation, training, and deployment. |
| Green-VLA | 2026.02 |
Method |
Temporal alignment, quality filtering, and embodiment-aware action interfaces. |
| LingBot-VLA | 2026.01 |
Code Weights Benchmark |
Efficient VLA post-training stack and GM-100 benchmark; raw pretraining data not confirmed open. |
| OpenVLA Data Tooling | 2024 |
Code |
OXE mixture conversion and fine-tuning utilities. |
| LeRobot Dataset Format | 2024 |
Code |
Standardized robot dataset format and tooling. |
| Hybrid-VLA Data Pipeline | 2025 |
Code |
RLDS conversion, action tokenization, and multimodal collation. |
| Qwen-VLA | 2026.05 |
Method |
Large-scale joint pretraining recipe across robot trajectories, egocentric demonstrations, synthetic simulation, VLN, trajectory supervision, and VLM data. |
| Work | Year | Artifact | Pipeline |
|---|---|---|---|
| ShareRobot | 2025 |
Data Code |
Task planning, affordance, trajectory, and QA annotations. |
| Robo2VLM | 2025.05 |
Data Code |
VQA generation from pose, gripper, force, and trajectory signals. |
| CAST | 2025.08 |
Method |
Counterfactual language/action relabeling that augments existing robot datasets with synthetic trajectory-instruction pairs. |
| Q-DIG | 2026.03 |
Method |
Quality-diversity prompt generation for VLA red-teaming; generated adversarial instructions are used to augment fine-tuning data. |
| CLIP-RT | 2024.11 |
Method |
Language-supervised data collection with stochastic trajectory augmentation and heuristic labeling for language-conditioned robot policies. |
| PixelVLA / Pixel-160K | 2025.11 |
Method Data |
Automated pixel-level annotation from robot data. |
| RoboVQA | 2023.11 |
Data Method |
Long-horizon robotics VQA and reasoning data. |
| RoboAfford++ | 2025 |
Method |
Generative affordance and spatial reasoning annotations. |
| Being-H0 | 2025.07 |
Method |
Human video, mocap, and VR data curation for dexterous VLA pretraining. |
| RLDX-1 | 2026.05 |
Code Weights Method |
Inverse dynamics labels for synthetic rare manipulation scenarios. |
| RoboInter | 2026 |
Method |
Intermediate representations and VQA/VLA-oriented robotic annotations. |
| Work | Year | Artifact | Pipeline |
|---|---|---|---|
| RoboGene | 2026.02 |
Data Method |
Agentic generation of diverse, feasible manipulation tasks. |
| RoboGene Dataset | 2026.02 |
Data |
Released RoboGene task-generation dataset. |
| LIBERO | 2023 |
Data Code |
Procedural language-conditioned task suites. |
| RoboCasa | 2024 |
Data Code |
Generated household tasks and environments. |
| RoboCasa365 | 2026.03 |
Data Method |
Large-scale household task and demonstration suite. |
| MolmoBot-Engine | 2025 |
Data Code |
Procedural task and scene generation. |
| ABot-N0 Data Engine | 2026.02 |
Method |
Expert trajectories and reasoning samples for embodied navigation. |
| RESample | 2025.10 |
Method |
Offline-RL critic and exploratory rollout sampling generate OOD/recovery data to improve VLA robustness. |
PRs are welcome. You can also send papers, corrections, artifact updates, or taxonomy suggestions to jake630@snu.ac.kr