Core Viewpoint - The article discusses the intersection of simulation technology, real data, and world models in the context of embodied intelligence, highlighting the ongoing debate about the importance of simulation versus real data and the potential breakthroughs in world modeling [3][11]. Group 1: Roundtable Discussion - The roundtable focuses on the "data dilemma" in embodied intelligence, featuring four young scientists who explore the boundaries between simulation and real interaction, as well as the technological advancements in world models like Genie [3][11]. - Sergey Levine's assertion that real data is irreplaceable is examined, questioning whether this is a strategic choice or an inevitable path in AI evolution [11]. Group 2: Key Participants - Li Hongyang, an assistant professor at the University of Hong Kong, leads the OpenDriveLab and has made significant contributions to end-to-end autonomous driving solutions, including the award-winning UniAD [4]. - Zhao Hao, an assistant professor at Tsinghua University, specializes in computer vision related to robotics and has co-founded over ten startups since 2009 [5]. - Gu Jiayuan, an assistant professor at ShanghaiTech University, focuses on generalizable robotic decision-making models and has received multiple awards for his research [6][7]. - Mu Yao, an assistant professor at Shanghai Jiao Tong University, has published extensively in top conferences and has received numerous academic honors [7].

Group 1: Industry Overview - The field of embodied intelligence is gaining momentum, with major tech companies globally investing heavily, resulting in billions in financing [1] - The World Robot Conference (WRC 2025) in Beijing showcased over 200 robotics companies demonstrating their capabilities, including various applications of embodied intelligence [1] Group 2: Understanding Embodied Intelligence - Embodied intelligence integrates AI into physical robots, enabling them to perceive and interact with the environment similarly to humans, learning through sensory feedback [2][4] - Non-embodied AI, or Internet AI, operates without physical interaction and relies on data input, contrasting with the experiential learning of embodied intelligence [2] Group 3: Data Challenges - The industry faces significant challenges in data acquisition, primarily due to high costs and the difficulty in generating large-scale datasets [5][7] - The need for high-quality, diverse data is critical, as embodied intelligence applications require extensive environmental data for effective operation [7][8] Group 4: Data Isolation and Solutions - The existence of "data silos" hinders data sharing between companies, leading to inefficiencies and wasted resources in the industry [8] - The reliance on synthetic data is increasing, with a significant portion of data in the embodied intelligence field being generated through simulation rather than real-world collection [9][10] Group 5: Future Prospects - The commercial viability of embodied intelligence robots is still in development, with mass production expected to take several more years due to high training and production costs [12] - The industry anticipates a future where embodied intelligence robots become commonplace in everyday life, although this transition may take time [12]

Group 1 - The core issue in embodied intelligence is the severe shortage of real data, with most robotic models relying on less than 1% of real operational data, which limits their generalization capabilities in complex environments [5][6] - There is a debate in the industry regarding the importance of real data versus synthetic simulation data, which affects the scalability and generalization of embodied intelligence [6][7] - Some experts argue that while synthetic data has advantages in cost and scalability, it cannot fully replicate the complexities of the real world, leading to a "domain gap" that hinders model transferability [7][8] Group 2 - The need for hundreds of billions of real data points is highlighted, with current datasets only reaching the million level, presenting a significant bottleneck for the development of embodied intelligence [8] - The strategy of using synthetic data for initial training followed by fine-tuning with real data is seen as a key pathway for the cold start and scaling of embodied intelligence [8][9] - Teleoperation is emerging as a primary method for acquiring real data, especially in the early stages of embodied intelligence, where human operators provide high-quality demonstration actions for training [9][10]

Core Viewpoint - Over-reliance on alternative data sources can severely limit the ultimate capabilities of models, and true breakthroughs must be built on real data [1][10] Group 1: The Dilemma of Alternative Data - Researchers in robotics often seek cheaper alternatives to real data due to high collection costs, leading to a compromise in model performance [2][3] - Common alternative methods include simulation training, learning from human videos, and using handheld devices to mimic robotic actions, but each method ultimately weakens the model's true potential [3][4] Group 2: Intersection Dilemma - The collection of data inevitably involves human judgment, which can limit the problem-solving approach when avoiding real data [4][6] - As models grow stronger, they can better distinguish between alternative and real data, leading to a smaller intersection of effective behaviors [6][7] Group 3: The Importance of Real Data - Attempting to bypass real data results in a "spork" scenario, where neither alternative data nor real data is effectively utilized [10][11] - To build robust robotic models that generalize well, real data is essential, but it can be complemented with diverse data sources [11][12] Group 4: The "Spork" Phenomenon - The concept of "spork" applies to various AI research areas, where attempts to combine manual design with learning systems ultimately create performance bottlenecks [13]