#United States #Austin #AI Robotics #Umi #Robotics Policy

Workshop Announcement: The UMI Arena

The AI Robotics Association (AIRoA), headquartered in Ota, Tokyo, is thrilled to announce that its workshop titled 'The UMI Arena: From Handheld Devices to Real-World Policies' has been accepted at the upcoming Conference on Robot Learning 2026 (CoRL 2026). This international conference is set to take place in Austin, Texas from November 9 - 12, 2026.

About the Workshop

The workshop aims to delve into the gathering of robot operation data utilizing handheld devices and the development of control policies derived from this data. Participants will be engaged in a collaborative environment where they can rigorously evaluate policies, data collection devices, and research outcomes based on common tasks and real-world conditions.

A pivotal component of this workshop will be the UMI Arena, an international competition where teams evaluate their developed policies in a standardized real-robot environment. To support model development effectively, AIRoA plans to share a portion of real-world robot operation data collected using their YUBI platform, alongside providing reference policy models and learning codes via a limited public access on GitHub.

Moreover, participants will benefit from hands-on demonstrations of UMI-type data collection devices, alongside a call for related research comprising a paper and poster session. This initiative aims to create an integrated forum where researchers and engineers worldwide can tackle shared challenges on robot operations.

Keynote Speakers

The workshop will feature renowned guest speakers at the forefront of robot learning and real-world robot operations, including:

• - Shuran Song from Stanford University
• - Andy Zeng from Generalist Robotics
• - Hao-Shu Fang from MIT CSAIL
• - Yang Gao from Tsinghua University

These speakers will address pressing topics regarding the UMI ecosystem, such as scaling demonstration data and the deployment of policies on robots with varied structures.

Context: The Expanding UMI Ecosystem

In the realm of robotics, significant operational data is essential for teaching robots diverse tasks. Traditional data collection approaches involving physical robots incur substantial costs in both equipment and operational logistics. In response to these challenges, the Universal Manipulation Interface (UMI) approach has garnered considerable attention. It allows the collection of human demonstration data without needing physical robots, using handheld devices.

Currently, there are several open-source devices available which enhance the efficiency of data collection. However, due to variations in robots, tasks, environments, and evaluation metrics employed across different research institutions, fairly comparing the performance of collected data and trained policies proves challenging. The UMI Arena seeks to standardize data collection, learning, and evaluation processes in real-world robot operations.

Workshop Objectives

• - Standardization in Real-World Robot Operation Evaluation: Aim to construct reproducible evaluation environments and protocols applicable to commonly available robotic hardware while ensuring sufficient complexity.
• - Policy Deployment Across Different Robots: Test how well policies trained based on hand manipulation adapt to varying tasks and robot models.
• - Device Evaluation for Data Collection: Discuss the implications of different device designs on data quality, operational capabilities, and cost.

Innovative Workshop Format

This workshop diverges from the traditional lecture format by incorporating live demonstrations, hands-on exhibitions, participant voting, poster sessions, and panel discussions. It invites participants to actively engage with speakers in addressing pressing challenges in the field.

Three Tracks of the UMI Arena

Track 1: Policy Competition

Teams are encouraged to submit pre-trained policy models to be evaluated within a standardized setup using common tasks and real-robot environments. This allows for the establishment of reproducible evaluation methods and shared benchmarks. Participating teams can include models such as Diffusion Policy and Vision-Language-Action (VLA) models.

Quality sessions with Q&A from top teams and live demonstrations using real robots will also take place, with a leaderboard maintained post-event.

Track 2: Data Collection Device Exhibitions

Participants can engage in hands-on sessions with UMI-type data collection devices, enabling them to operate and assess these tools up close. Exhibitors will present their devices along with examples of policies derived from the operational data collected, promoting evaluation based on ease of use, performance, and cost.

Track 3: Papers and Poster Sessions

Research on UMI-based robot operation learning is welcomed, with selected papers presented in an interactive format alongside some spotlight talks. Themes include policy deployment across different robots, data quality issues, robust localization, haptic sensing, and scalable learning.

Awards will be given for the 'Best Policy', 'Best Device', and 'Best Paper' per track, encouraging innovation and excellence.

Continuous Contributions to Robotics

The UMI Arena Workshop seeks not only to yield results from presentations and competitions but also to ensure benefits for the community through:

• - An evaluation foundation for robot operation policies based on shared real-world environments
• - Reproducible evaluation protocols and public leaderboards
• - Common evaluation metrics for device comparison
• - Insights into policy deployment across varying robots
• - Reusable real-world datasets, baseline models, and learning codes
• - An international community of developers and researchers focusing on device development and policy research.

Upcoming Events

The tentative schedule includes:

• - Post Acceptance Announcement: Registration details, baseline details, and tool releases.
• - Mid-August 2026: Initial data availability for participants.
• - September 2026: Gradual rollout of public data leading to approximately 20,000 hours of data.
• - October 2026: Deadline for policy and paper submissions.
• - November 12, 2026: Workshop take-off and announcement of winners.

Stay tuned for the official page for further updates on the schedule.

Workshop Overview

• - Title: The UMI Arena: From Handheld Devices to Real-World Policies
• - Conference: Conference on Robot Learning 2026 (CoRL 2026)
• - Date: November 12, 2026
• - Location: Austin, Texas, USA
• - Type: Half-day in-person event
• - Special page: The UMI Arena
• - Organizer: AI Robotics Association (AIRoA)

Upcoming seminars or submitted speaker presentations, data content releases, and deadlines may be updated on the dedicated webpage.

In conclusion, the UMI Arena promises to be a compelling opportunity for professionals in the field of robotics to share, learn, and advance the state of robot operation learning in the real world.