RLCAD: Reinforcement Learning Training Gym for Revolution Involved CAD Command Sequence Generation

Xiaolong Yin1*, Xingyu Lu1*, Jiahang Shen1, Jingzhe Ni1, Hailong Li2, Ruofeng Tong1, Min Tang1, Peng Du1†,
1Zhejiang University 2Shenzhen Poisson Software Co., Ltd.
(* denotes equal contribution, † denotes the corresponding author)
Paper Code(Coming soon) Dataset(Coming soon)

Face-Extrusion and Revolution Operation

Battery modeling process using the gym interface. The leftmost image shows face IDs on the surfaces. The right sequence illustrates four extrusion/revolution operations with Boolean operations applied iteratively to generate the final geometry.

Network Architecture

CAD-MLLM

The training pipeline is composed of two stages. In the first stage, a contrastive learning approach is employed to pre-train the UV-Net network, aiming to derive an encoder model that can effectively characterize the B-Rep of the CAD model. During the second stage, a reinforcement learning approach is employed to generate the command sequence. We first utilize the tunable UV-Net model to extract the B-Rep embedding of the CAD model, which is then integrated with the feature vector of the historical modeling action sequence. Subsequently, the Actor-Critic network predicts the action distribution and value. The predicted action is transmitted to RLCADGym for execution, yielding the next-stage observation. The neural reward and geometric reward are designed to update the policy network.

Our Dataset(Omni-CAD)

Result Comparison of Reconstruction

(Please view our paper for more results)
CAD-MLLM

Comparison of generation results with Fusion 360 Gallery. It shows our method generates higher-quality results in terms of both completeness and detail.

CAD-MLLM

Comparison of generation results with cadrille and CAD-Recode. It shows our method generates higher-quality results in terms of both completeness and detail.

BibTeX

If you find this work useful, please cite:


@article{yin2025rlcad,
  title={Rlcad: Reinforcement learning training gym for revolution involved cad command sequence generation},
  author={Yin, Xiaolong and Lu, Xingyu and Shen, Jiahang and Ni, Jingzhe and Li, Hailong and Tong, Ruofeng and Tang, Min and Du, Peng},
  journal={Computer-Aided Design},
  pages={104027},
  year={2025},
  publisher={Elsevier}
}