Robot Era's robot foundation model is a brand-new learning paradigm native to AI, combining ultimate generality and simplicity, and fully capable of scaling up. Its algorithm processes visual, auditory, and tactile sensory information through multimodal input fusion, recreating the most fundamental interaction between organisms and the physical world. This original design not only restores the essence of natural interaction but also enhances cross-modal information processing capabilities, allowing intelligent agents to perceive and understand the environment more comprehensively and flexibly. In terms of generality, Xingdong Jiyuan adheres to the "the bitter lesson" philosophy, enabling robots to abandon human prior knowledge, and achieve true autonomous evolution and widespread application through self-exploration and self-learning. In terms of simplicity, the company uses an end-to-end architecture, allowing intelligent agents to adapt to tasks and environments in real-time without complex programming or human intervention, significantly improving flexibility and development efficiency.

STAR 1 has already adopted an end-to-end pure learning method in movement and operation, relying solely on a neural network to achieve generalized walking and grasping in different environments. Its cutting-edge language-vision-action large model generates actions directly from language and visual signals, avoiding information transmission loss and supporting more efficient scaling-up. After extensive data training, the model has developed excellent generalization capabilities, allowing the robot to master new skills with minimal data in new task scenarios, demonstrating superior adaptability and extensibility.