AI★★★★arXiv · 2026-07-02
Controllable Sim Agents with Behavior Latents
A new framework, Controllable Neural Variational Agents (CNeVA), is proposed to learn behavior latents from agents and make their behavior controllable. This controllability enables engineers to isolate variables, reproduce specific edge cases, and test autonomous systems without real-world risk. The CNeVA framework learns a per-agent Gaussian behavior latent from per-channel discounted returns.
📌 Key points
- CNeVA learns a per-agent Gaussian behavior latent
- The framework makes agent behavior controllable
- Controllability enables isolation of variables and reproduction of edge cases
本页为 gitzw.com 基于公开来源的 AI 中文解读，非原文转载。