Michael C. Welle is a Postdoctoral Researcher working with Danica Kragice in EECS/RPL at KTH Royal Institute of Technology focusing on representation learning for deformable object manipulation. He obtained his MSc in Systems, Control and Robotics at KTH in January 2018. His subsequent Ph.D. research was performed under the supervision of Danica Kragic and the Co-supervision of Anastasia Varava and Hang Yin resulting in his successful defense with the title “Learning Structured Representations for Rigid and Deformable Object Manipulation” in December 2021.
Speaker Links: Website
The performance of learning based algorithms largely depends on the given representation of data. Leveraging representations in deformable object manipulation has received increased interest from the community in recent years.
In this talk, I will argue that more beneficial representations can be obtained by structuring the representation space. I will introduce our work regarding the Latent Space Roadmap (LSR) framework, which structures the latent space using a contrastive loss term and performs a subsequent clustering step in order to build a graph directly in the latent space that is used for visual action planning. Furthermore, I will introduce our current work regarding the Augment-Connect-Explore (ACE) paradigm that addresses the issue of data scarcity in Visual action planning.
Papers covered during the talk
- Lippi, Martina, et al. “Latent space roadmap for visual action planning of deformable and rigid object manipulation.” 2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2020.
- Lippi, Martina, et al. “Enabling Visual Action Planning for Object Manipulation through Latent Space Roadmap.” Conditional accepted at TRO 2021, arXiv preprint arXiv:2103.02554 (2021).
- Lippi, Martina, et al. Augment-Connect-Explore: a Paradigm for Visual Action Planning with Data Scarcity .” Submitted to International Conference on Intelligent Robots and Systems (IROS) 2022