PROFACTOR is a research and technology company focusing on industrial inspection, robotic/digital assistance, micro/nanomanufacturing, and digital printing technologies. We research and develop the production of the future. Our technological developments optimize the production of renowned industrial partners.

Manipulation of deformable objects such as textile garments remains one of the major challenges in robotics due to high variability in shape, occlusions, wrinkles, and non-rigid dynamics. Reliable grasp point estimation is a crucial step toward enabling autonomous robotic systems to handle clothing for applications such as flattening, folding and sorting.

This thesis focuses on AI-based grasp point estimation for category-level garment manipulation. The objective is to predict meaningful grasp locations on deformable garments such that a robot can manipulate the cloth and lay it flat on a table or conveyor for downstream tasks including folding, cloth-type classification, and automated handling. The work will investigate how dense visual correspondence methods can be used to transfer grasping knowledge from known garments to previously unseen garments of similar categories.

A potential direction is to leverage human demonstrations or expert-defined reference grasp points and learn correspondence-aware models that can estimate equivalent grasp locations on unseen garments.

The research will particularly explore category-level generalization, where the system learns manipulation-relevant visual features independent of garment instance variations. The thesis will be built upon recent state-of-the-art (SotA) research works such as UniGarmentManip [1] and DexGarmentLab [2]. The project may involve simulation-based experimentation as well as evaluation on robotic manipulation benchmarks.

[1] R. Wu, H. Lu, Y. Wang, Y. Wang, and H. Dong, “Unigarmentmanip: A unified framework for category-level garment manipulation via dense visual correspondence,” in IEEE/CVF Conference on Computer-Vision and Pattern Recognition (CVPR), pp. 16340–16350, June 2024.

[2] Y. Wang, , et al., “Dexgarmentlab: Dexterous garment manipulation environment with generalizable policy,” in Neural Information Processing Systems (NeurIPS), 2026.

What to expect

The master thesis comprises the following tasks:

• Familiarize with the literature on deformable object grasping and manipulation
• Evaluate existing SotA pretrained models from UniGarmentManip and DexGarmentLab for generalization capability on unseen garments and configurations.
• Design and implement a custom grasp prediction framework using modern learning approaches such as diffusion-policy-based methods, or implicit neural representations / feature fields.
• Compare the proposed approach against the SotA baselines using quantitative and qualitative evaluation metrics.
• Workload split: Research 30%, programming and implementation 50%, and writing 20%

Your profile

• A committed student who works independently and meticulously, and who is enthusiastic about theory and its practical application.   
• Comfortable working with AI tools; pursuing studies in automation engineering, electrical engineering, mathematics, artificial intelligence, and other technical disciplines.

We offer

• An interesting and well-defined thesis topic within an ongoing R&D project
• The opportunity to test results in a practical setting
• A well-equipped robotics laboratory
• Support from highly engaged colleagues

Start/ Duration

Starting as soon as possible – Duration of 4 months (extendable depending on performance)

We offer for master thesis a compensation of min. 550 EUR per month.

We are looking forward to your application – please apply at www.profactor.at

Contact

Questions? Please contact: Alexandra Siegl, asiegl@profactor.at, +43 7252 885 225.