The embodied intelligence track is experiencing a milestone breakthrough. Recently, ForceGPT officially launched three core products of its Embodied Native series, among which the most notable to the industry is the world's first embodied native large model DM0. This not only marks that robots now have an "original" foundation for their brains, but also indicates that embodied intelligence will accelerate into real productivity.

"Embodied Native" Starting from Zero: Giving Robots Intuition

Different from previous approaches of rewriting large language models, DM0 is an embodied native large model trained from scratch by ForceGPT and Step星辰.

Deep Integration: The model integrates multimodal internet information with multisensory data from embodied scenarios, achieving the unification of perception and action.

Multi-task Coverage: The pre-training phase combined three core tasks, and it has already successfully covered eight mainstream robot models.

Strong Generality: It has excellent cross-machine generalization and migration capabilities, meaning developers can quickly adapt one logic to different hardware forms.

2.4B Parameters Achieve "A Small Weight Moves a Big One", Real Machine Evaluation Ranks First Globally

In an era where parameter count is not everything, DM0 proves the value of efficiency. The 2.4B parameter version won two first places in the RoboChallenge real machine evaluation, currently ranking first globally. This performance, achieving a lot with little, greatly reduces the reliance of robots on backend computing power.

Comprehensive Open Source: Consumer-level Graphics Cards Can Also Handle Robot Development

ForceGPT announced that DM0 has been fully open-sourced, which is undoubtedly a big gift for global developers.

Developer-friendly: Developers can even fine-tune and redevelop the model on consumer-level graphics cards, greatly lowering the innovation threshold.

Ecosystem Closure: Along with the release, the embodied native development framework Dexbotic 2.0 and the embodied native application mass production workflow DFOL were also launched, building a complete chain from development to mass production.

Industry Insight: The "Android Moment" of Embodied Intelligence?

With the open source of DM0 and the implementation of the embodied native framework, robot development is moving from "craft workshops" to "industrial mass production." When a large model with strong cross-machine generalization capability becomes infrastructure, the scale of the explosion of embodied intelligence will far exceed our imagination.