Competition Introduction

With the rapid development of the embodied intelligence sector, intellegent algorithms related to Imitation Learning (IL) , Reinforcement Learning (RL), and large-scale decision-making models have already achieved substantial advances in theoretical foundations. However, the push of evolution of algorithms towards practical usability and high robustness still requires continuous real-world interaction and feedback processes, and such process is faced with several resource-straining barriers:

• High cost of entry: High-performance robotics hardware still carry an astronomically expensive price tag, ranging from hundreds of thousands to even millions of dollars.

• High dataset production costs: High-quality datasets require complex synchronisation systems and manual annotations on specialised facilities, which incurs high costs and long development cycles

• Lack of a standardised testing platform: There lacks a standardised testing platform, which greatly increases the difficulty of comparing algorithms in a fair manner under real-world conditions.

Being faced with these bottlenecks, this brand-new competition promises the following:

• First-in-world to release large datasets collected from our fleet of real twin-armed humanoid robots from specialised facilities

• Fully open-sourcing an evaluation platform with real-world robots, along with support for participant’s remote deployment and evaluation of their algorithms

• Provides various self-testing systems developed in-house, with standardised interfaces for all the tasks, as well as a full technical support community.

Our goal is to lower the technical barrier of entry, so that every participant can efficiently access their required resources, have a quick onboarding experience, and have easy access to algorithm validations.