arXiv cs.AI· Yufeng Peng, Yixin Zhang, Ziyi Liu, Zhikun Xu, Jiarong Yang, Sijie Wang, Wenjie Feng, Qiyu Kang·· 2 天前AI 评分31
DeciCAD：为 CAD 智能体打造基于 harness 的动作策略学习
DeciCAD: Harness-Grounded Learning of Action Policies for CAD Agents
AI 导读
DeciCAD 是一个面向 CAD 领域的原生 harness，通过暴露语义 CAD 动作、决策导向观测与环境反馈，从交互轨迹中提取因果决策监督，并用 SFT 与几何接地的在线 RL 训练动作策略。
正文 · 原文
Abstract:Large Language Model agents operate through closed-loop action policies whose decision space is defined by an agent harness. We investigate how this operating space shapes agent behavior and policy learning in computer-aided design (CAD), a professional domain characterized by persistent state, geometric constraints, and long-horizon procedures. We instantiate this perspective in DeciCAD, a domain-native harness that exposes semantic CAD actions, decision-oriented observations, environment feedback, and persistent state transitions. Building on this harness, we develop a learning pipeline that extracts causal decision supervision from interaction trajectories and improves the policy through supervised fine-tuning (SFT) and geometry-grounded online reinforcement learning (RL). Across four CAD benchmarks, DeciCAD achieves the best F1 and intersection over union (IoU) while reducing model-output tokens by 45--88% relative to a general-purpose coding harness. Under matched supervision, its SFT policy improves F1 from 0.6883 to 0.7437 and outperforms the coding policy; subsequent online RL further raises geometry availability from 66.5% to 73.5% and F1 to 0.7880. These results demonstrate that the decision space induced by a harness materially affects both agent execution and policy learning in CAD.
| Subjects: | Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Machine Learning (cs.LG); Robotics (cs.RO) |
| Cite as: | arXiv:2610.21755 [cs.AI] |
| (or arXiv:2610.21755v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2610.21755 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Yufeng Peng [view email]
[v1]
Thu, 15 Oct 2026 12:11:43 UTC (3,078 KB)
来源：arXiv cs.AI · arxiv.org