DeciCAD: Harness-Grounded Learning of Action Policies for CAD Agents

Key points

• Large Language Model agents operate through closed-loop action policies whose decision space is defined by an agent harness.1
• We instantiate this perspective in DeciCAD, a domain-native harness that exposes semantic CAD actions, decision-oriented observations, environment feedback, and persistent state transitions.1
• 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).1
• These results demonstrate that the decision space induced by a harness materially affects both agent execution and policy learning in CAD.1

Sources (1)

• [1]DeciCAD: Harness-Grounded Learning of Action Policies for CAD Agents
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 15, 12:11 PM

  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.

  Large Language Model agents operate through closed-loop action policies whose decision space is defined by an agent harness.

Extractive summary: sentences quoted from the sources.