手机+机械手 人人都能“训练”机器人

Core Insights - The article discusses the importance of data in training embodied intelligence and introduces the RoboPocket solution by Qunche Intelligent, which allows ordinary people to contribute to data collection in real-world settings [1][2]. Group 1: RoboPocket Solution - Qunche Intelligent has launched the RoboPocket solution, which utilizes smartphones, apps, and lightweight mechanical hands for high-quality data collection in various real-life scenarios [1][2]. - The RoboPocket system aims to democratize data collection, allowing individuals to gather valuable data from their homes, thus expanding the data collection process beyond traditional data factories [2]. Group 2: Data Collection Methodology - The RoboPocket system includes a two-finger mechanical hand and a smartphone that together form an intelligent data collection module, capable of real-time environment mapping and action guidance [2]. - Users can perform everyday tasks such as folding towels or organizing snacks, which can be converted into learnable signals for robots, ensuring data quality through immediate feedback [2]. Group 3: User Participation and Incentives - The solution lowers the barrier for participation, enabling ordinary individuals to become data contributors and potentially receive rewards for completing data collection tasks [4]. - The initiative aims to create a positive feedback loop where everyone can participate, data becomes more diverse and valuable, and models can evolve accordingly [4]. Group 4: Future Developments - Qunche Intelligent plans to release the RH20T dataset in 2023 and the CoMiner field companion collection system in 2025, further enhancing the data collection ecosystem [4]. - The company envisions a future where various professions and daily habits serve as potential learning materials for robots, enriching the training data available for embodied intelligence [4].