Human-to-Robot Handovers

Yes. Teams can request to use one of the available robotic arms (UR5 or Franka Emika Panda) at the competition site in Vienna. However, availability is limited, and teams that bring their own setup will be prioritised. Please indicate your preference during registration as early as possible to improve allocation chances.

Teams bringing their own equipment are responsible for logistics and transportation to Vienna. We recommend arranging early with ICRA 2026 organisers about venue access for equipment setup. The on-site robotic arms may be available for teams with significant logistical constraints. Indicate this preference during registration.

The complete list of required equipment includes:

• Robotic arm (6+ degrees of freedom, e.g., UR5, KUKA) or request to use provided robot
• End effector (e.g., Robotiq 2F-85 parallel gripper, or equivalent)
• Laptop/workstation (tower PC with monitor)
• Cameras with tripods and USB-C cables (up to 2, e.g., Intel RealSense D435i)
• Digital scale (to weigh containers before and after)
• Calibration pattern/board (for camera calibration)
• Power extension leads and socket adapters
• Ethernet cable for PC and robotic arm connection

Equipment shipping and advance storage must be coordinated with ICRA 2026 organisers. We recommend contacting them directly about logistics, venue access for setup, equipment delivery deadlines, and storage facilities. Visit https://2026.ieee-icra.org/ for venue and logistics contact information. More details will be provided also by the competition organisers in the following weeks.

No. The task must be performed autonomously without any human input or teleoperation. Your solution should make all decisions and control the robot without manual intervention.

No. You cannot use prior knowledge of the objects (e.g., prior 3D object models or reconstructed container shapes from previous handovers). The only prior knowledge available is the high-level, unspecific set of container categories: cup, drinking glass, and food box.

Yes. You are allowed to use existing Large Language Models and APIs as part of your solution. However, you must ensure that your system can function without internet connectivity, as WiFi at the competition site may be unstable and unreliable. We recommend designing your solution to work offline for robustness.

You can use any combination of sensing modalities including:

• RGB cameras
• RGB-D cameras (depth sensors)
• Microphones (for sound-based detection)
• Tactile sensors
• Force/torque sensors

Multimodal approaches are encouraged and allowed. Just ensure all sensors are properly calibrated and synchronised.

Yes, this is allowed. A system using one RGBD camera mounted on the robot is an acceptable alternative approach to using two regular cameras, provided it captures the necessary data for your solution and meets calibration requirements.

Before executing any configurations, teams must:

1. Synchronise all cameras and verify they record simultaneously
2. Calibrate cameras (both intrinsic and extrinsic parameters)
3. Test and verify the behaviour of the robotic arm (end-effector, speed, kinematics, etc.)
4. Have a digital scale ready and calibrated
5. Test communication between PC and robotic arm over ethernet
6. Conduct safety checks and have a safety procedure ready

The robotic arm must have:

• At least 6 degrees of freedom
• 2-finger parallel gripper (e.g., Robotiq 2F-85 or equivalent)
• Adequate reach to grasp containers from the centre of the table
• Repeatability and reliability for executing handover tasks
• Verified kinematics and speed behaviour

You may use either approach:

• External fixed cameras (2 RGB cameras recommended): Provides wide field of view but requires careful positioning
• Robot-mounted camera (1 RGBD camera): More compact, moves with robot end-effector, but requires adequate cable management

Both are acceptable; choose based on your solution design.

Teams must prepare the sensing setup such that the cameras are:

• SYNCHRONISED: Cameras must record in synchronisation with each other
• CALIBRATED: Intrinsic and extrinsic parameters must be determined
• LOCALISED: Camera position and orientation must be known relative to a calibration board/pattern

This ensures accurate 3D reconstruction and localisation of containers during handover.

We recommend the cameras record RGB sequences at:

• Frame rate: 30 Hz (frames per second)
• Resolution: 1280 × 720 pixels

These specifications are based on the setup used in the CORSMAL Benchmark protocol. However, you may use different specifications if they suit your solution—document any deviations in your submission.

The two cameras (or single RGBD camera) should be placed:

• Distance: 40 cm from the robotic arm (e.g., using tripods)
• Orientation: Both positioned to view the centre of the table
• Height: Adjust for optimal visibility of hand-object interactions
• Lighting: Position to minimise glare and shadows

The setup typically requires a space of approximately 4.5 × 4.5 meters.

No. The set of objects for the on-site competition consists of 6 unknown containers that are revealed only on competition day. These containers test your system's generalisation capabilities by presenting novel and unseen objects. The containers may range between drinking cups and food containers, potentially filled with different content and amounts than used in the qualification phase.

Yes. Be aware of:

• Illumination: Lighting conditions at the competition venue may differ from your laboratory. Design your perception system to be robust to varying light conditions.
• WiFi: WiFi on the conference site might be unstable and unreliable. Do not rely on internet calls for critical functionalities.

If a handover fails (e.g., the robot cannot grasp and/or hold the container during delivery, or the object falls after placement), that configuration receives 0 points. You can attempt other configurations, but you cannot retry a failed configuration. Document failures carefully in your submission.

Yes. Previous competition results provide insights into typical performance levels and approaches:

(1) RGMC 2024 ICRA track: https://corsmal.github.io/rgm24icra/index.html

(2) RGMC 2025 Handover track: https://corsmal.github.io/rgmc2025-handover-track/

These resources show winning teams' approaches and performance metrics.

Teams must participate in the qualification phase by performing the CORSMAL Benchmark protocol handover configurations remotely in their laboratory. Results are submitted to the organisers, and the top-performing teams are ranked on a leaderboard. Teams qualify for the on-site competition based on performance and resources available (robots). A maximum of 6 teams will qualify.

The deadline for the qualification phase is 16 April 2026.

The full CORSMAL protocol must be executed (i.e., all 288 handover configurations) and results must be submitted to us so that we can compute the scores and update the leaderboards. The full protocol ensures fair comparisons and comprehensive evaluation of your solution across all scenarios.

Results must be submitted so that we can compute all the scores and update the CORSMAL leaderboards. Submitting partial results is possible but can lead to lower benchmark scores. The leaderboards are designed to enable fair comparisons and provide results for research labs worldwide to reuse for experiments and publications. Note that the evaluation protocol for the on-site competition is simplified and does not compute all the properties evaluated in the full CORSMAL Benchmark.

The s8 score measures how accurately your robot reaches 6 standard target poses in its own workspace. You submit the reached end-effector poses (position + orientation) and your robot's workspace parameters, and the score is computed from the normalised position error using a 3 cm threshold. Full instructions, scoring details, and submission templates are available in the s8 documentation.

No. You do not need to purchase exact containers from commercial links. You can prepare equivalent containers that closely match the specified physical properties (shape, size, weight, stiffness). Please provide a document describing your container specifications including dimensions, material, weight ranges, and filling options and content types. We will review and confirm if they meet the requirements for fair comparison.

The DOI link (https://doi.org/10.17636/101CORSMAL1) should work correctly. If you experience issues, try these alternatives:

(1) Direct QMUL repository: https://qmro.qmul.ac.uk/xmlui/handle/123456789/103670

(2) Detailed dataset information: https://corsmal.github.io/containers_manip.html

(3) Benchmark webpage: https://corsmal.github.io/benchmark.html

If problems persist, contact Dr. Changjae Oh ([email protected]) directly. Note: You do not need to complete the CORSMAL Benchmark before accessing the dataset.

Dates, deadlines, fees, and other registration information for ICRA 2026 can be found on the official ICRA 2026 registration information page: https://2026.ieee-icra.org/attend/registration-information/

All teams competing on-site must have registered for the conference. We recommend registering for the conference early as it may be required for visa applications.

The Human-to-Robot Handovers competition will be held in Vienna, Austria during ICRA 2026:

• Conference Dates: 1-5 June 2026
• Venue: Vienna, Austria
• Venue Details: Messe Wien Exhibition & Congress Center

Detailed venue information and logistics: https://2026.ieee-icra.org/attend/venue-and-logistics/

ICRA 2026 is held in Vienna, which offers excellent travel connections and accommodation options. For travel planning and accommodation:

• Visa Information: Register for ICRA 2026 early—a competition pass may be required for visa applications
• Travel: Vienna has excellent connections via Schwechat Airport and train stations
• Accommodation: Visit the ICRA 2026 venue page for official partner hotels and recommendations

Contact ICRA 2026 organisers for specific travel support queries: https://2026.ieee-icra.org/

Yes. Teams are invited to contribute a research article on their solution, findings, and results to the Topical Collection on "Advancing Robotic Grasping and Manipulation for the Real-World" within Autonomous Robots: https://link.springer.com/collections/ifcjfbbgji

This is an excellent opportunity to disseminate your research to the academic community.

Beyond publication, you have opportunities to:

• Submit a demo video to be showcased on the RGMC YouTube channel
• Present your work at the competition in Vienna
• Contribute to the CORSMAL leaderboards and benchmark database
• Network with other researchers and roboticists
• Receive recognition and prizes (top-performing teams)
• Obtain a free MATLAB license from Mathworks to support development

Yes, demo videos are optional but strongly encouraged. Demo videos provide excellent visibility for your work and team. Videos will be showcased:

• During the competition in Vienna
• On the RGMC YouTube channel, reaching the research community worldwide
• In competition materials and documentation

While specific technical requirements will be provided, we recommend:

• Duration: up to 10 minutes (recommended: 2-5 minutes)
• Content: Show your approach, key components, and live demonstrations
• Quality: HD resolution (1080p or higher)
• Format: MP4 or similar widely compatible format
• Audio: Clear narration explaining your solution

Videos should demonstrate your system's capabilities and approach to the handover problem. See RGMC YouTube channel for examples.

Qualified teams can submit demo videos by 26 May 2026. The final deadline for submitting videos for the RGMC YouTube channel is 30 June 2026. Submit videos directly to Dr. Changjae Oh ([email protected]).

For technical questions during the qualification phase, contact:

Dr. Changjae Oh
Queen Mary University of London
Email: [email protected]

He can provide guidance on:
• Benchmark execution
• Data and dataset access issues
• Protocol clarifications
• Setup and calibration questions
• Submission procedures

Please contact the track organisers directly:

Dr. Changjae Oh
Queen Mary University of London
[email protected]
He will be happy to answer your questions about the competition.

Visit the official CORSMAL ICRA 2026 website: https://corsmal.github.io/events/rgmc/icra2026/

For general RGMC competition information: https://sites.google.com/view/rgmcomp