Emerging Learning and Algorithmic Methods for Data Association in Robotics

8:25-14:50 EDT, Sunday, May 31

Summary:

- The workshop was held virtual on Zoom, Sunday, May 31, 2020, with over 400 registrants.

- We thank all participants, invited, and spotlight talk speakers.

- The talks are available to public at: https://youtu.be/-lEVvtsfz0I

Outline:

This workshop aims to present the latest results and emerging learning and algorithmic techniques for data association in robotics. Data association, which can be described as identifying relations between sets of measurements, physical objects, labels, etc., is a well-studied field with solutions dating back to 70’s. However, recent results in the fields of optimization, graph theory, and machine learning have opened new and exciting research directions. Through a series of contributed and invited talks by academic leaders and renowned researchers, emerging algorithmic methods based on optimization or graph-theoretic techniques, learning and end-to-end solutions based on deep neural networks, and the synergy between them will be discussed. These techniques promise new capabilities and performance improvements across a broad range of applications, including but not limited to semantic segmentation, point cloud alignment, sensor fusion for autonomous vehicles, object pose estimation for robotic manipulation, place recognition for simultaneous localization and mapping, and data fusion for multi-agent systems. The workshop will further facilitate discussion on current challenges and research directions in the next 5-10 years.

Expected outcome:

By facilitating discussion among participants, authors of contributed papers, and invited speakers, the workshop aims to study and answer the following fundamental questions:

1) What are the latest results and emerging directions for formal methods such as optimization and graph-based techniques?

2) What are the latest results and emerging directions for end-to-end solutions and learned models?

3) Can these methods work in synergy and how to combine mathematical and learned models?

4) How to utilize these methods in real-time systems with limited computational resources or in distributed systems with limited communication capacity and what are the trade-offs?

5) What are the correct notions, quantifications, and measures of accuracy for data association?

6) How is "failure" defined in data association and can a failed association still be useful?

Dissemination:

We cordially invite researchers to submit short papers, extended abstracts, or reports. Submitted contributions can describe work in progress, preliminary results, novel concepts, and applications in industry. All manuscripts are limited to 4 pages and should use the IEEE standard two-column conference format (paper template available on the IEEE ICRA 2020 website). We encourage authors to submit a video clip to complement their manuscript. Submissions will be selected based on their originality, relevance to the workshop topics, contributions, technical clarity, and presentation. All accepted manuscripts will be presented as posters during three poster sessions that are spread out throughout the workshop schedule. The top four contributions will be given 15 minutes to present their work during two spotlight sessions. Topics of interest include

Methods for outlier rejection and resilient data association
Optimization and relaxation techniques for data association
Machine learning methods with neural or statistical models
Methods based on synergetic mathematical and learned models
Point cloud alignment, sensor registration, fusion, and map merging
Semantic segmentation, object detection, and pose estimation from images or point clouds
Applications of perception and data association algorithms in autonomous vehicles, robotic manipulation, localization, and mapping

To submit your contributions please follow: https://cmt3.research.microsoft.com/EDAT2020

Please contact the corresponding organizer with any questions.