| --- |
| pretty_name: EvenFlow Benchmark |
| license: other |
|
|
| task_categories: |
| - robotics |
| - computer-vision |
|
|
| task_ids: |
| - trajectory-prediction |
| - motion-planning |
|
|
| tags: |
| - navigation |
| - human-robot-interaction |
| - trajectory-data |
| - multi-agent |
| - benchmarking |
|
|
| annotations_creators: |
| - no-annotation |
|
|
| source_datasets: |
| - original |
|
|
| language: |
| - en |
|
|
| size_categories: |
| - 1K<n<10K |
| --- |
| |
| # EvenFlow Benchmark Dataset |
|
|
| EvenFlow is a shared-space navigation evaluation suite built from real-world human trajectory data. |
|
|
| Most benchmarks evaluate whether an agent can navigate *around* people. |
| EvenFlow evaluates whether an agent can navigate *with* them. |
|
|
| This repository contains the benchmark data used to construct executable navigation tasks: |
| - layouts |
| - scenes |
| - tasks |
| - human trajectory tracks |
|
|
| Code and evaluation tools are available at: |
| https://github.com/standard-ai/evenflow-benchmark |
|
|
| --- |
|
|
| ## Dataset Structure |
|
|
| ``` |
| benchmark/ |
| layouts/ |
| aligned_flow/ |
| cross_flow/ |
| interaction_constrained/ |
| ``` |
|
|
| --- |
|
|
| ## Responsible AI Considerations |
|
|
| ### Data Collection |
| Data was collected in real-world environments using overhead camera systems. |
| Only anonymized trajectory data is released. |
|
|
| ### Privacy |
| No raw video, biometric identifiers, or personally identifiable information are included. |
|
|
| ### Intended Use |
| This dataset is intended for research on: |
| - navigation in human environments |
| - multi-agent coordination |
| - trajectory-based evaluation |
|
|
| ### Limitations |
| - Single-site data collection |
| - No demographic attributes |
| - Offline evaluation setting |
|
|
| --- |
|
|
| ## License |
|
|
| This dataset is released under a custom license. |
|
|
| - Free for research and non-commercial use |
| - Commercial use requires a separate agreement |
|
|
| See LICENSE for details. |
|
|
