Skip to main content

Towards Efficient and Portable Robot Learning for Real-World Settings

Track: Robotics and AI / Data

Date and Time: 14 Mar 2023 08:30-09:50

Organisers: Riccardo Zanella, Roberto Meattini, Néstor García, Ashok Sundaram

Main questions to be answered: How can we take advantage of robotic priors, scene structure, and demonstrations to accelerate robotic learning and minimize the reliance on real-world data in the learning process?

Additional questions: How can control theory be integrated into the learning framework to enforce system theoretic properties?

Additional questions 2: How can we ensure safety when the robot agent needs to physically explore an unknown environment?

Workshop Description: The next generation of robotic manipulation systems will witness an increase in skills thanks to novel AI paradigms, enabling them to handle complex tasks in unstructured environments and adapt to unexpected circumstances. For autonomous robotic systems to be relevant in practical real-world applications, the learning process must be both data efficient and safe, leveraging on a priori knowledge and models about the environment and tasks. This includes the vital ability to transfer skills between applications and robotic systems while ensuring constant safety under all conditions. The workshop seeks to address critical questions at the heart of efficient and portable robot learning for real-world applications. By bringing together experts and enthusiasts, our objectives are as follows: * Harnessing Prior Knowledge: Explore the use of robotic priors, scene structures, and demonstrations to accelerate the learning process. * Integration of Control Theory: Discuss how control theory can be integrated into the learning framework to ensure system-theoretic properties are maintained. * Efficiency and Skill Acquisition: Delve into methods for robots to efficiently acquire the skills necessary for high-performance manipulation. * Safety and Exploration: Consider strategies to ensure the safety of robotic agents when they venture into unknown environments. * Reducing Data Dependence: Examine innovative strategies to minimize the reliance on real-world data in the learning process. The expected outcome of this workshop is the generation of innovative ideas, collaborations, and practical solutions. Participants will leave with a deeper understanding of how to accelerate robot learning, integrate control theory, ensure safety, and reduce data reliance. This knowledge will drive advancements in robotics for real-world applications. We will make the workshop accessible to a wider audience by posting workshop recordings on our website for future reference. Additionally, the insights and findings from the discussions will be shared in bite-sized updates, ensuring that valuable information is easily digestible and readily available to the robotics community.

Intended Outcome: Workshop discussion topics of common interest, success stories, use cases, etc, Workshop expected to create a roadmap or white paper

Approach: From 08:35 to 09:15, attendees will engage with Invited Speaker’s Presentations, during which Dr. Sylvain Calinon, Dr. Maximo A. Roa, Prof. Angela Schoellig, and Prof. Georgia Chalvatzaki will provide insights into their research topics. Following these presentations, a panel discussion will take place from 09:15 to 09:50. Furthermore, from 11:40 to 12:04, participants will have the opportunity to attend an Insight Session featuring posters showcasing accepted contributions.

Contributors: * Dr. Sylvain Calinon, Senior Research Scientist at the Idiap Research Institute [Confirmed] * Prof. Georgia Chalvatzaki, Full Professor for Interactive Robot Perception & Learning at the Technical University of Darmstadt [Confirmed] * Dr. Maximo A. Roa, Senior Scientific Researcher at the German Aerospace Center – DLR [Confirmed] * Prof. Angela Schoellig, Alexander von Humboldt Professor for Robotics and Artificial Intelligence at the Technical University of Munich, Associate Professor at the University of Toronto Institute for Aerospace Studies [Confirmed]

Further Information: