CloudGripper: The Open Source Cloud Robotics Testbed

IEEE ICRA 2026 Cloud Robotics Competition

Thanks to all the teams that participated in the IEEE ICRA 2026 Cloud Robotics Competition track which was part of the 11th Robotic Grasping and Manipulation Competition at IEEE ICRA 2026.

Teams focueds on developing robust manipulation solutions using remote access to the cloud robotics platform CloudGripper (cloudgripper.org) at KTH Royal Institute of Technology, where a set of small robotic arm cells were made available for competing teams for 10 weeks to develop and evaluate their solutions. The 2026 cloud robotics track focussed on generalization and robustness properties of the developed solutions. Teams worked on two foundational manipulation tasks: a) planar pushing of a variety of objects and b) shape control of linear deformable objects with varying physical properties. The goal will be to not only solve these tasks, but to propose methods that can be applied also when variations of physical properties are introduced.

2026 Competition Winners

  • 1st place: Team K-DAS
    Sumin Kim, Jinyoung Jeong, Bohyeong Kim, Kiheon Kwak, Suchul Kim
    Kookmin University
  • 2nd place: Team MIL-Cloud
    Shenghan Huang, Mingwei Ma, Jiahao Long, Jiawei Weng, Yang Cong
    South China University of Technology
  • 3rd place: Team PulseBot
    Kaiji Zeng, Liduo Hu, Qingxi Hu, Wei Liu
    South China University of Technology

Competition Timeline

  • Jan 19th, 2026
    Announcement of competition details.
  • Extended to: March 16th, 2026
    Registration deadline. Complete this form to apply as a team or individual participating in the competition.
  • March 2nd, 2026
    Kick-off workshop with registered teams to prepare for remote access to CloudGripper.
  • March 16th - May 22nd, 2026
    During these 10 weeks, teams are given remote access to CloudGripper robots in pre-arranged booking slots to develop solutions. Teams will use their own local coding environment in Python and will use a remote access API is used to send commands and receive video data from the robots.
  • 1st June - 5th of June
    IEEE ICRA. During the competition time slot, a final evaluation of the teams will be conducted.

Newsletter, Signup and Contact

More information about this competition track is expected to be released in early 2026. Contact us at info (at) cloudgripper (dot) org to notify us about your interest and we will send a reminder when the competition details are announced.

The CloudGripper Testbed

A preprint about CloudGripper is available here: "CloudGripper: An Open Source Cloud Robotics Testbed for Robotic Manipulation Research, Benchmarking and Data Collection at Scale", IEEE ICRA 2024, Muhammad Zahid and Florian T. Pokorny. Teams will be given access to a simple position control API with a Python interface to remotely work with the robots.

Competition Task 1: Planar Pushing

This task will focus on the planar pushing of objects to a target location. During the initial preparation phase, teams will be given access to a set of sample objects and remote robot work cells using which algorithms can be developed. We will focus on the following key points to evaluate performance, generalization and robustness of the developed methods:

  • The competition is agnostic to the type of method you want to develop. Both Foundation Models and Classical Methods are welcome in particular.
  • Developed solutions will have to perform well on any of the CloudGripper robot cells under minor variations in scene appearance and calibration/noise differences.
  • Some objects will have an un-even internal weight distribution.
  • The same algorithm should work for any of the considered objects.
  • Previously unseen objects will be introduced in the final evaluation.
  • Solutions will be evaluated using an integral of intersection-over-union calculation between current and target pose over time. As a result, it will be important to develop solutions that can quickly solve the pushing challenge. A time limit per trial will also be applied.

Competition Task 2: Linear Deformable Object Shape Control

This task will focus on controlling the shape of a linear deformable object (like a rope) to match a target pose. During the initial preparation phase, teams will be given access to robot cells with a set of sample linear deformable objects using which algorithms can be developed. Analogously to the first task, we will focus on evaluating overal performance as well as generalization and robustness of the developed methods:

  • Developed solutions will have to perform well on any of the CloudGripper robot cells under minor variations in scene appearance and calibration/noise differences.
  • The same algorithm should work for any of the linear deformable objects to be considered.
  • Previously unseen linear deformable objects with differing physical properties will be introduced in the final evaluation.
  • Solutions will be evaluated using an integral over time of a shape matching metric. As a result, it will be important to develop solutions that can quickly solve the challenge. A time limit per trial will also be applied.