AGIBOT World Challenge at ICRA 2026 Introduces New Tracks for Robotics Innovators
The AGIBOT World Challenge has officially opened for ICRA 2026, setting the stage for a thrilling competition that aims to push the boundaries of robotics. With a prize pool of $530,000, participants will engage in two distinct tracks: the Reasoning to Action Track and the World Model Track. This year’s event promises to build on the successes of last year’s inaugural challenge, driving innovation in the field of embodied intelligence and robotics.
This track evaluates the capability of models to reason and execute actions effectively, integrating both online simulations and real-robot stages. Competitors will be tasked with training AI models capable of addressing complex challenges using the AGIBOT WORLD open-source dataset. The focus here is to bridge the gap between simulation and real-world applications, achieving robust and effective interactions in diverse environments.
Participants will work on tasks designed to reflect real deployment scenarios across logistics, industrial, retail, and domestic settings, making the competition practical and relevant. Key tasks include:
As part of the competition, AGIBOT provides a specially curated dataset, which is available for download to ensure participants have the resources needed for success. They’ll also receive access to ACoT-VLA, the baseline model, which will help them navigate the training and evaluation workflow effectively. AGIBOT’s Genie Sim 3.0 will be utilized for comprehensive model evaluation, integrating 3D reconstruction and visual generation.
The World Model Track shifts the spotlight onto the essential capability of embodied world models. Here, participants will train models for accurate video generation of robotic actions based on real observations and command signals across various environments. The key aim is to model physical dynamics accurately based on robot interactions, incorporating a further understanding of potential failures during operations.
With 30,000 real trajectories amassed from 10 tasks, including various interaction types like grasping and moving, the quality of generated interactions will be assessed meticulously. Participants can base their entries on the EVAC model, AGIBOT’s proprietary world model that has been pre-trained for high efficiency. The objective is to achieve the innovative application of generative models in robotics.
The competition timeline is set to keep the momentum going. Key milestones include:
As these dates approach, teams from all around the globe are encouraged to join this exciting opportunity to showcase their skills and creativity in robotics. Throughout the challenge, AGIBOT will facilitate a community space for ongoing discussions, technical support, and updates.
For more detailed information about the challenge and registration, participants can visit AGIBOT's official website. The future of robotics innovation is bright, and AGIBOT is at the forefront of this remarkable journey, showcasing what the next generation of robotics can achieve.
Reasoning to Action Track
This track evaluates the capability of models to reason and execute actions effectively, integrating both online simulations and real-robot stages. Competitors will be tasked with training AI models capable of addressing complex challenges using the AGIBOT WORLD open-source dataset. The focus here is to bridge the gap between simulation and real-world applications, achieving robust and effective interactions in diverse environments.
Participants will work on tasks designed to reflect real deployment scenarios across logistics, industrial, retail, and domestic settings, making the competition practical and relevant. Key tasks include:
- - Continuous sorting
- - Barcode scanning
- - Logistics sorting
- - Shelf stocking
- - Door opening
As part of the competition, AGIBOT provides a specially curated dataset, which is available for download to ensure participants have the resources needed for success. They’ll also receive access to ACoT-VLA, the baseline model, which will help them navigate the training and evaluation workflow effectively. AGIBOT’s Genie Sim 3.0 will be utilized for comprehensive model evaluation, integrating 3D reconstruction and visual generation.
World Model Track
The World Model Track shifts the spotlight onto the essential capability of embodied world models. Here, participants will train models for accurate video generation of robotic actions based on real observations and command signals across various environments. The key aim is to model physical dynamics accurately based on robot interactions, incorporating a further understanding of potential failures during operations.
With 30,000 real trajectories amassed from 10 tasks, including various interaction types like grasping and moving, the quality of generated interactions will be assessed meticulously. Participants can base their entries on the EVAC model, AGIBOT’s proprietary world model that has been pre-trained for high efficiency. The objective is to achieve the innovative application of generative models in robotics.
Competition Flow
The competition timeline is set to keep the momentum going. Key milestones include:
- - Global registration opened on February 12, 2026
- - Access to competition servers commenced on February 28, 2026
- - Final submissions deadline on April 20, 2026
- - Online stage results to be announced by April 30, 2026
- - The offline finals will take place at ICRA 2026 starting June 1, 2026.
As these dates approach, teams from all around the globe are encouraged to join this exciting opportunity to showcase their skills and creativity in robotics. Throughout the challenge, AGIBOT will facilitate a community space for ongoing discussions, technical support, and updates.
For more detailed information about the challenge and registration, participants can visit AGIBOT's official website. The future of robotics innovation is bright, and AGIBOT is at the forefront of this remarkable journey, showcasing what the next generation of robotics can achieve.
Topics Consumer Technology）
【About Using Articles】
You can freely use the title and article content by linking to the page where the article is posted.
※ Images cannot be used.
【About Links】
Links are free to use.