Zhi Tong Cai Jing·2025-05-16 05:52Core Insights - The hardware maturity of humanoid robots is currently higher than that of software, with software being the key to commercialization [1] - Tesla's Optimus is focusing on algorithm improvements, highlighting the importance of controllers, operational technology, chips, and data collection equipment [1] Algorithms: The Core of Embodied Intelligence - The algorithm framework is divided into two levels: the upper "brain" focuses on task-level planning and decision-making, while the lower "cerebellum" handles real-time motion planning and joint control [2] - The upper control utilizes visual-language-action (VLA) models for semantic understanding and action generation, while the lower control is transitioning from traditional model-based control to modern reinforcement learning (RL) and imitation learning (IL) [2] - Future breakthroughs in algorithms need to address challenges such as multi-modal integration, long-term task planning, and Sim-to-Real transfer [2] Data: The Foundation of Algorithm Learning - The quality and diversity of data directly impact algorithm performance, with three main sources: real data (highest precision but lowest volume), synthetic data (cost-effective but domain gap issues), and web data (large scale but requires cleaning) [3] - Real data is primarily collected through teleoperation and motion capture technologies, while synthetic data is generated via simulation platforms like NVIDIA Omniverse [3] Control Systems: The Foundation of Embodied Intelligence - There is no consensus in the industry regarding the division of humanoid robots into "brain" and "cerebellum," with the brain handling complex algorithms and data processing, and the cerebellum managing motion control [4] - The hardware consists of SoC chips, including CPU, GPU, and NPU, along with processors, storage units, communication interfaces, and I/O interfaces [4] - The software includes the operating system, middleware, and upper-layer software, with chips being the core of the controller, often utilizing NVIDIA solutions [4] Investment Opportunities in the Industry - Software is the focus for the next phase of commercialization in robotics, with Tesla's Optimus showing marginal changes primarily in algorithms [5] - Key investment targets include: - Controllers: Tianzhun Technology (embodied intelligence controllers), Zhiwei Intelligent (humanoid robot-specific controllers), Desay SV (leading vehicle domain controller) [5] - Operational technology: Huichuan Technology (PLC and drivers), Xinjie Electric (PLC), Leisai Intelligent (PC-based controllers and boards), Gokong Technology (PC-based controllers and boards), Tuosida (industrial controllers) [5] - Chips: Rockchip (SoC chips), Horizon Robotics (robotics layout) [5] - Data collection equipment: Lingyun Optical (optical motion capture equipment), Aofei Entertainment (stake in Noyton, optical-inertial integrated motion capture solutions) [5]