Lingjing: A Simulation Platform for Multi-Agent Embodied Intelligence in Open-Ended Cities
A novel simulation platform named Lingjing has been launched to enhance urban embodied intelligence, necessitating collaboration among diverse agents like UAVs, ground robots, and autonomous vehicles within dynamic urban settings. This platform, discussed in a paper on arXiv (2608.08045), overcomes the shortcomings of current simulators that separate various embodiments and detach them from task design and assessment. Lingjing reconstructs and visualizes evolving cities using geographic data, synchronizes multiple physics engines, and provides agents with access to a shared physical and structured urban state. It includes a Gym-like interface for user-defined ReAct agents and supports both single and multi-agent natural-language missions, featuring configurable communication methods and resource constraints. Each episode is an attribution-ready replay linking agent trajectories and communications. The paper was submitted to arXiv, identified as 2608.08045v1.
Key facts
- Lingjing is a simulation platform for heterogeneous multi-agent embodied intelligence in open-ended urban environments.
- It reconstructs and renders evolving cities from geographic data.
- It synchronizes multiple physics engines.
- It exposes shared physical and structured urban state to agents.
- It has a Gym-like interface supporting user-defined ReAct agents.
- It supports single- or multi-agent natural-language missions.
- Communication can be star or broadcast with resource constraints.
- Each episode is attribution-ready, linking agent trajectories and communication.
- The paper is available on arXiv with identifier 2608.08045v1.
Entities
Institutions
- arXiv