Embodied Intelligence GPT-2 Moment Arrives: A Chinese Company Creates the World’s LargestEnd-to-End Unified Embodied Large Model
Recent reports on Physical Intelligence (PI), a company backed by OpenAI, have highlighted the transformative potential of embodied intelligence large models in ushering in a new era of robotics. While the focus has been oninternational advancements, a domestic company in China, X Square, has emerged as a pioneer in this revolution.
X Square, a startup, is the only companyin China that has embraced the end-to-end unified large model technology route from day one, aligning with PI’s approach. Their Great Wall operation large model series, WALL-A, is even surpassing PI in terms of parameter scale.
In April, Machine Intelligence reported on a demo by X Square showcasing their self-developed end-to-end unified embodied large model. This model enabled a dual-arm robot to perform intricate tasks on irregular objects, suchas grasping, picking, cutting, folding clothes, and making beverages, using low-cost hardware. This demonstrated a significant level of generalization capability.
Currently, X Square’s model has reached impressive levels of performance, particularly in handling long-sequence complex tasks, generalization, and versatility. For instance, zippinga zipper, a notoriously challenging task for robots, requires more than just positional control. The robot needs to accurately align the slider with the teeth, apply the right amount of force, and adapt to the deformation of the fabric in real-time. It also needs to accurately judge the zipper’s engagement status and handle anomalieslike fabric getting stuck.
This breakthrough signifies the arrival of the embodied intelligence GPT-2 moment. X Square’s achievements demonstrate the potential of embodied intelligence models to revolutionize the robotics industry.
This article is based on information from Machine Intelligence and other reliable sources. The author has conducted in-depth researchand analysis to ensure accuracy and originality.
References:
- Machine Intelligence: [Link to the original article]
- Physical Intelligence: [Link to PI’s website or relevant news article]
- X Square: [Link to X Square’s website or relevant news article]
Views: 0