HA-VLN Documentation
HA-VLN provides human-aware navigation data and environments for studying instruction-following agents in scenes with dynamic humans.
This documentation focuses on the public HA-VLN workflow:
- installing the HA-VLN runtime and
HASimulator - preparing the public HA-VLN data package
- developing and testing agents against the human-aware environment
- understanding the simulator APIs and human-aware metrics
If you are getting started, you usually need to answer three questions:
- How do I install
HASimulatorand obtain the public data? - Which APIs can I call when building or adapting an agent?
- After an agent can run, how do I test it on public splits?
This documentation is organized around exactly those three tasks.
Recommended Reading Order
1. Install HASimulator and Prepare Data
Start here if you are setting up the environment for the first time.
2. Develop an Agent with HASimulator APIs
Read this when you already have the environment running and want to build your own method.
3. Test an Agent
Read this after your agent can run and you want to validate it on public HA-VLN data.
Reference Material
If you need more detail, the reference pages are still available:
- installation details: Dependencies, Installation Steps, Data Download
- integration details for local method development: Agent Integration Notes
- simulator and metric APIs: Human State, Scene Updates, Collision Checks, Evaluation Metrics