3D CovDiffusion

3D-CovDiffusion is a model designed for 3D coverage path planning, with a focus on robotics applications. The tool provides checkpoints for category-specific pretrained policies, enabling the application of diffusion-based approaches to coverage tasks in 3D environments. It is associated with diffusion-policy methods and operates on 3D point cloud data. The available resources for 3D-CovDiffusion include released model checkpoints, which contain EMA weights from specific training runs, exact inference configurations, and provenance metadata. These checkpoints are organized by categories such as windows, cuboids, shelves, and containers, each corresponding to different datasets and training runs. The tool is implemented using PyTorch and the safetensors format. Additionally, a train-ready dataset for 3D-CovDiffusion is provided, allowing users to further train or evaluate the model as needed. The intended use is for coverage path planning in 3D spaces, particularly in robotics contexts. Delivery is through downloadable model files and datasets hosted on Hugging Face, with supplementary code and project documentation available via linked project and code pages. Information about licensing or specific pricing is not provided in the available evidence.