A grand challenge for robotics is generalization; to operate in unstructured real world environments we need household robots that can quickly learn to perform tasks in unseen kitchens, mobile manipulators and drones that can navigate novel spaces, and autonomous vehicles that can safely maneuver through unseen roads with varying conditions, all while minimizing dependence on humans. Recent breakthroughs in natural language processing and vision suggest that the secret to this level of generalization is data — not just the amount of data collected, but its diversity, and how to best leverage it while learning. Especially challenging is that most large-scale sources of data are collected offline, from varied sources. How do we use this diverse, offline data to build generalizable robotic systems?
Speakers
Call for Papers
L-DOD @ ICRA 2023 seeks high-quality research papers that introduce new ideas and stimulate future trends in robotics and machine learning. We invite submissions in all areas of data-driven robot learning and machine learning, including but not limited to:
Diverse dataset collection, benchmarking, offline RL, imitation learning, transfer learning, pretraining/finetuning, cross-embodiment learning, generalization challenges and real-world applications requiring diverse datasets such as manipulation, navigation, and autonomous driving.
This year, we are specifically soliciting submissions around the following discussion areas:
Data Sourcing and Priors
What type of data is most conducive to learning meaningful robot priors?
Collecting and sourcing offline video data (humans)
Collecting and sourcing offline video data (robots)
Impact of multimodal data (language/narrations plus videos, audio)
On the need for "diversity" in data – what this means and why it's important
The complexity of inductive biases and strong priors contained within internet-scale dataset
Learning Control End-to-End from Offline Data
How might we learn end-to-end behavioral policies for robotics from offline data?
What changes when we’re looking at different sub-areas of robotics? Handling long-term dependencies? Alternate sensor modalities?
Deploying these models efficiently on robots
Should we learn end-to-end control, or is learned perception/language understanding + classical controllers enough?
Using Large-Scale Data for Perception, Planning, and More!
Incorporating offline data at various points in the robotics stack (e.g., for perception, planning & reasoning, environmental priors)
Active data selection (how to select relevant data for downstream labeling)
Handling multi-embodiments present in the data, sub-optimality
What should be the role of large pretrained internet-scale models?
Many recent works aim for open-ended decision making. How much progress has the field of robot learning made towards this, and what bottlenecks remain?
Note: These topics are not exhaustive! If you feel your work fits with the spirit of this workshop, we heartily encourage you to submit!
Submission Guidelines
Submission Portal: OpenReview
Paper Format: Submissions can follow the template recommended by any major conference in robotics and machine learning (RSS/CoRL/NeurIPS/ICML/ICRA/…). We are not enforcing a strict template and the authors are free to choose a template of choice, as long as the paper clearly indicates that it was a submission for the Workshop on Learning from Diverse, Offline Datasets at ICRA 2023 .
Double Blind Review: All submissions to the workshop must be properly anonymized for double-blind review. Any author information or information that may otherwise identify author identities should be removed.
Page Limit: Following the ICRA 2023 guidelines, there is a 6 page limit for technical content with no page limit on references and appendices. We encourage relevant works at all stages of maturity, ranging from initial exploratory results to polished full papers.
LLM Usage Policy: We allow the use of LLMs for paper writing as long as it's clearly stated how the LLM was used and which parts of the paper were influenced in any way by the LLM. The full policy is here.
Dual Submission:
Papers to be submitted or in preparation for other archival venues (e.g. IROS, CoRL, NeurIPS, ISRR etc.) in the field are allowed (as long as this is OK with the archival venue in question!)
We also welcome recently accepted or published works (e.g. RSS, ICML), but this must be explicitly declared at the time of submission.
Visibility: Submissions and reviews will not be public. Only accepted papers will be made public via the workshop’s OpenReview page.
This workshop is a non-archival venue and will not have official proceedings. Workshop submissions can be subsequently or concurrently submitted to other venues.
For questions or concerns regarding the peer review process, please contact the workshop organizers at ldod_icra2023@googlegroups.com
Location
The workshop will be hybrid with the in-person venue at South Gallery Room 19
Organizers
Code of Conduct
L-DOD 2023 is organized for the purpose of open exchange of ideas, the freedom of thought and expression, to engage in productive debates, create professional connections, and learn about exciting research.
L-DOD 2023 is committed to ensuring all participants have a positive experience at the symposium. All participants have the equal right to pursue shared interests without harassment or discrimination in an environment that supports diversity and inclusion. L-DOD will not tolerate any harassment, discrimination, personal attacks, disruption, or bullying. Participants who are asked to stop such behavior are expected to do so immediately.
If you have concerns about a participant's behavior, please reach out to ldod_icra2023@googlegroups.com or contact any of the organizers. We will respond as soon as possible.