SOCIA-$\nabla$: Textual Gradient Meets Multi-Agent Orchestration for Automated Simulator Generation

TL;DR

SOCIA-$ abla$ introduces an innovative framework that combines textual gradients with multi-agent orchestration to automate simulator generation, achieving state-of-the-art accuracy across multiple CPS tasks with minimal human effort.

Contribution

The paper presents SOCIA-$ abla$, a novel end-to-end agentic framework that unifies multi-agent orchestration with textual gradient optimization for automated, reproducible simulator code generation.

Findings

Achieves state-of-the-art accuracy on three CPS tasks.

Unifies multi-agent orchestration with loss-driven optimization.

Scales across domains and simulation granularities.

Abstract

In this paper, we present SOCIA-$\nabla$, an end-to-end, agentic framework that treats simulator construction asinstance optimization over code within a textual computation graph. Specialized LLM-driven agents are embedded as graph nodes, and a workflow manager executes a loss-driven loop: code synthesis -> execution -> evaluation -> code repair. The optimizer performs Textual-Gradient Descent (TGD), while human-in-the-loop interaction is reserved for task-spec confirmation, minimizing expert effort and keeping the code itself as the trainable object. Across three CPS tasks, i.e., User Modeling, Mask Adoption, and Personal Mobility, SOCIA-$\nabla$ attains state-of-the-art overall accuracy. By unifying multi-agent orchestration with a loss-aligned optimization view, SOCIA-$\nabla$ converts brittle prompt pipelines into reproducible, constraint-aware simulator code generation that scales across domains and simulation granularities. We will release the code soon.