Argus improves reliability of simulated users for training agents
ARGUS: An Agenda-Driven Framework for Reliable User Simulation
Artificial Intelligence
Summary
Simulators that act like real users help teach and test computer programs that interact with people, but current ones often miss goals or don’t follow rules well. The authors introduce ARGUS, a system that keeps track of user goals separately from language, ensuring it sticks to plans and constraints. They also created a way to train ARGUS efficiently and measure how well it behaves. Tests on popular benchmarks showed ARGUS performs better than existing simulators.
What this means in practice
- •For dialog system developers: Create more reliable user simulators to train dialog agents with strict adherence to user goals and constraints.
- •For conversational ai trainers: Use ARGUS-based training to improve agent evaluation through faithful user interactions without extra costly human labeling.
Authors
Hang Zheng, Nan Jiang, Zichen Zhu, Zixu Yang, Zhiyang Tang, Shuai Fan, Lu Chen, Xiaobao Wu, Kai Yu
Abstract
High-fidelity user simulators are essential for evaluating and training agents in interactive environments. However, current LLM-based simulators struggle with reliability issues, such as goal omission and policy non-compliance. We propose ARGUS (Agenda-dRiven Goal-oriented User Simulator), a neuro-symbolic framework that maintains the user's goal state using a dynamic agenda stack manipulated via symbolic operations. By rigorously decoupling state tracking from natural language generation, ARGUS facilitates strict goal progression and constraint-aligned interactions. For cost-efficient deployment, we distill ARGUS into compact models using a unified pipeline: evaluator-guided SFT followed by online RL with dense step-level rewards. To rigorously evaluate behavioral fidelity, we introduce a rubric-based metric Composite Behavioral Score (CBS). Evaluations on $τ^3$-Bench and MultiWOZ demonstrate the significant superiority of the ARGUS architecture and the efficacy of our training pipeline. Code and data will be released soon.