Pudu Robotics, the global leader in commercial service robots with over 130,000 units shipped and a 23% market share, has launched two foundational technologies for embodied intelligence: PuduFM 1.0, a robot foundation model, and PuduAgent, a general-purpose embodied agent platform.
The centerpiece is a strategy called "One Brain, Multiple Embodiments", where robots of different forms—delivery, cleaning, industrial AMRs, the D5 robot dog, the D7 semi-humanoid robot, and the D9 full-size humanoid—all share one unified, end-to-end model and software architecture.
PuduFM 1.0 introduces an industry-first Physical Intuition Model (PIM) built on a Causal-Attention Transformer, enabling robots to predict physical outcomes before acting. A child running into a robot and tipping its coffee cup? PIM reads the risk instantly and adjusts posture to steady the tray. The companion Vision-Language-Action (VLA) module achieves deep alignment of language, vision, and action in a unified feature space for the first time.
PuduAgent provides a three-layer architecture—System, Capability, and Safety—with Agent Core for long-horizon task decomposition, Agent Memory for persistent learning, and cross-robot capability sharing. The platform includes an SDK, simulation environment, and SkillHub marketplace for third-party developers.
