About
The workshop titled "Learning Dexterous Manipulation" aims to investigate learning-based approaches for dexterous manipulation with a high level of generalizability. Dexterous manipulation has been one of the most challenging problems in robotics, and this workshop intends to offer insights and perspectives to researchers and participants on this topic. Additionally, the latest advancements in various sensing technologies will also be discussed. The ultimate goal of the workshop is to equip participants with the knowledge and skills necessary to design and develop advanced robotic systems capable of performing complex manipulation tasks with perception and enhancing human-robot interaction and collaboration.
This workshop is intended for researchers, engineers, and students who have a solid background in learning-based approaches, computer vision, or other related fields, and are interested in robotics and robot sensing. The presenters and panelists for the workshop will include experts from both academic and industrial backgrounds, representing a variety of disciplines, such as robot learning, robotics, mechanical engineering, and robot sensing. Accepted papers will have a chance to be presented during the poster session, and selected papers will be featured in contributed talks. The workshop will be promoted through relevant mailing lists of universities and research institutes, as well as social media platforms. Here are the topics we are interested in, covering recent advancements and open questions in the context of learning dexterous manipulation.
- Data for Dexterous Manipulation:
- Can human hand data for dexterous manipulation be collected in a general way using any expert-grade equipment? What is the data gap between human and robot hands?
- How can we improve current data collection methods, such as teleoperation, to facilitate large-scale data collection?
- Computer Vision:
- How can occlusion between objects and robot hands during dexterous manipulation be addressed?
- How can policies generalize to the open world outside the lab environment, considering the relatively unpredictable changes in outdoor lighting and the vast amount of information that needs to be processed?
- Tactile Information:
- How can tactile information help robots better accomplish tasks and perceive their environment?
- What kind of tactile information is best suited for dexterous robot hands, and can it compensate for the shortcomings of visual perception?
- Robot learning
- Will we see a unified and generalized model for most daily dexterous manipulation tasks or a specialized model for each individual task?
- How can learning-based policies handle dynamic tasks that require high-frequency control and detailed dynamics models?
