PuduFM 1.0 is an embodied AI model from Pudu Robotics

Pudu Robotics has unveiled PuduFM 1.0, an embodied AI foundation model that gives robots a common cognitive system to operate in the real world.

Architecture and Key Features

The model integrates movement planning and task execution into a single continuous decision loop. Navigation and manipulation are processed together rather than as separate steps.

PuduFM 1.0 is based on two key layers:

Physical intuition model—analyzes object behavior and constraints.

The multimodal vision-language-action system turns this analysis into action.

This architecture takes robots beyond simple object recognition. They begin to predict outcomes, understand contact and resilience, and choose actions with foresight.

Pudu positions this system as a step towards universal physical agents that can adapt to different environments and tasks. With a Humanoid Guide Brain Score of 6, PuduFM 1.0 is in the “capable and mature” range, reflecting a software stack that is ready for practical deployment.

Training pipeline and data

The main technical advantage of PuduFM 1.0 is its training pipeline, which combines large-scale data from real deployments with simulation and human feedback.

Pudu says the model learns from over 130,000 deployed robots in over 80 countries. This gives it a broad operating base, covering a variety of environments and use cases.

Training in a high-fidelity simulator speeds iterations, and refinement with live human feedback helps align decisions with practical results in the field. This makes the system relevant for robots that need:

understand the behavior of objects;

anticipate the consequences of contact;

coordinate movement with action in a single process of reasoning.

The same basic intelligence is used in delivery platforms, cleaning platforms, industrial robots and humanoid robots. This allows you to use accumulated knowledge about the world instead of building autonomy from scratch each time.

Strategy and market implications

PuduFM 1.0 is listed at zero dollars and is available to order - a signal that it is positioned as a platform smart offering rather than a conventional consumer product.

Its broader significance stems from the "One Brain, Many Incarnations" strategy, which allows different forms of robots to share a common and constantly improving model. This approach:

shortens development cycles;

disseminates training throughout the park;

creates a more powerful data accumulation cycle as more robots work in more places.

By building on progress in deployed systems rather than just laboratory demonstrations, Pudu links fundamental model development to commercial robotics at scale. This is important for the entire market because it points to software-defined robotics, where improvements in world modeling and action policy could improve the utility of entire fleets of robots in services, industry and humanoid robots.