ANet4D Documentation
ANet4D: simulation infrastructure for collaborative embodied agents. ANet4D aims to be for collaborative embodied-agent research what NS-3 is for network research: a shared experimental base in which teams of embodied and software agents are studied in one world, on one clock and under one evaluation protocol. The intended scope covers multi-UAV fleets, unmanned ground vehicles (UGV), unmanned surface vessels (USV), ground robots, fixed sensors and software agents without a body.
The current foundation is what V0.1 to V0.3 delivered:
Real-world 4D digital twins. Captured or reconstructed city point clouds and photogrammetric meshes become versioned Worlds that serve physics, queries and display, with a 4D environment field (wind, visibility, precipitation) on the same simulation clock.
Interactive Sandbox (V0.2). One isolated simulation core per visitor with fleet simulation, targets and perception quantification, nests and tasking, a REST interface and a Python SDK; the server needs no GPU and the scene renders in the browser.
Event Lab (V0.3). Event tapes compiled from declarative event specifications, Gazebo Sim lockstep as the high-fidelity physics tier, mesh-rendered camera, thermal and LiDAR sensors, and a benchmark runtime with an agent SDK and baseline policies; sensor generation requires an NVIDIA GPU on the server.
The delivered entity type today is the drone (multirotor, fixed-wing and VTOL airframes). The forward direction is a cross-entity collaborative experimentation platform with declarative experiment specifications, replaceable collaboration algorithms and comparable evidence; see Versions and roadmap.
This documentation covers 0.3.4 (V0.3.4 · ANet4D). The public instance runs at anet4d.agentnetwork.org.cn, and the source code is on GitHub. ANet4D was previously published as ANet Drone4D; the former site and documentation addresses redirect to the new ones, and a small set of legacy identifiers keeps the former name (see Frequently asked questions).
Getting help
When you run into a problem, look in this order:
Read the Frequently asked questions and the troubleshooting section of Quick start.
Search or open an issue on GitHub Issues, in English or Chinese.
For changes that affect a contract (file formats,
awr.rt.v1, the REST interface, reason codes), open an issue for discussion first; see CONTRIBUTING.md.Report security issues to hi@anet0.com; see SECURITY.md.
First steps
- Overview
Positioning, core capabilities and the delivery status of each (delivered, in development, research).
- Installation and requirements
Server and browser requirements, ports, directories, environment variables and run profiles.
- Quick start
From
git cloneto a local shared show and a sandbox session, then create a session, add a drone, read its state and end the session with the Python SDK.
User guide
- Shared show
The read-only shared show: scenario S7, cameras and sensor views, targets and relay statistics; each of the other six cities plays back its own recorded monitoring demonstration.
- Sandbox
One isolated sandbox per visitor: worlds and scenario templates, management panels, clock and rate, manual control, snapshots and world switching, lifetime and power saving.
- Targets and perception
Target classes, attributes and sensitivity ranges, the discovery criterion, the detection, recognition and identification levels, and definite capture.
- Tasks, nests and relay
Six task types and area guard, nests and batteries, GSD scan planning, 7x24 relay monitoring and sustainability assessment.
- Phones and tablets
Device classes, layouts, touch gestures and quality settings on phones and tablets.
- Internal identity entry
The identity entry for internal staff: where it is and what it is for.
- ANet4D Python SDK quickstart
The
anet4dPython SDK of the Event Lab (V0.3): install it, write an example agent and run a 3-drone episode on the high-fidelity and fast simulation tiers.
Open interface
- Open interface
Endpoint tables of the sandbox REST v1 and the local runtime API (generated from the OpenAPI snapshots), authentication, reason codes, quotas and the realtime protocol.
- Python SDK
The single-file Python SDK
awr_sandbox: installation, object model, call reference, error handling and examples.
System and data
- Architecture and coordinate frames
Process topology, data flow, clocks and the sandbox layer; conventions for World ENU, body FLU, the render frame and the PX4 boundary.
- Worlds and data
World Package v1, the seven worlds, the synthetic demo city, and the authorization and citation of UrbanScene3D.
- Deployment
The deployment kit of the public demo, resource limits, path routing and how this documentation site is published.
All the rest
- Versions and roadmap
What V0.1 to V0.3 delivered, the scope of V0.3.2 and V0.3.3, and the forward direction of ANet4D (D4).
- Acceptance summary
Results of the second D2 acceptance round, root causes of the failed items, and links to the full reports.
- Frequently asked questions
Frequently asked questions.
- Citation and license
Citing ANet4D and UrbanScene3D, license and third-party notices.