Back to Ecosystem Pulse
ToolProduction Ready
virtualhome
by xavierpuigf
Unity-based multi-agent household simulator with programmable agent control
Python
Updated May 20, 2026
Share:
How It Works
Provides an API to run VirtualHome, a Unity-based multi-agent household simulator for embodied AI research. Exposes scene control, object interaction, and agent action scripting so you can run reproducible multi-agent scenarios and collect observations and logs. task definitions Includes task definitions and programmatic control that make it easy to instrument agent interactions and failure cases.
Why It Matters
As agents interact in shared physical environments, reproducible simulation is essential to evaluate agent-to-agent behavior and failure modes. VirtualHome makes it possible to run controlled multi-agent experiments, reproduce scenarios, and gather the signals needed for agent track records and A2A evaluation. This kind of grounded testing is a prerequisite for meaningful reputation or continuous evaluation systems before deploying agents in the real world.
When to Use
Researchers and engineers who need a reproducible, embodied embodied simulator to test multi-agent interactions, delegation, and failure modes in household domains.
Use Cases
- Run reproducible multi-agent household scenarios to evaluate agent interactions and delegation strategies
- Collect detailed interaction logs and observations for building agent track records or reputation datasets
- Stress-test agent failure modes and recovery behaviors in a controlled embodied environment
Works With
unitypythonreinforcement-learningcomputer-vision
Topics
computer-visiondeep-learninggraphmulti-agentreinforcement-learningsimulatorunity
Similar Tools
ai2thorhabitat
Keywords
multi-agentsimulatoragent-evaluationembodied-ai