软体机械臂也能“精准操控”？新型神经网络控制策略破局！

Core Viewpoint - The article discusses the advancements in soft robotics, particularly modular soft robotic arms (MSRA), which address the limitations of traditional rigid robotic arms in complex, unstructured environments. The introduction of a bi-directional Long Short-Term Memory (biLSTM) network for intelligent control enhances the flexibility and adaptability of MSRA in various applications such as minimally invasive surgery and disaster rescue [1][2][22]. Group 1: Soft Robotics Development - Traditional rigid robotic arms dominate industrial automation due to their high precision and load capacity, but they face limitations in safety, environmental adaptability, and human-robot interaction in complex tasks [1]. - Soft robotics, based on biomimicry and flexible materials, offers unique advantages such as continuous deformation, high collision safety, and strong environmental compliance, opening new application dimensions [1]. Group 2: Modular Soft Robotic Arms (MSRA) - MSRA is a core research direction that decomposes robotic arms into standardized, interchangeable functional units, providing reconfigurability, scalability, and ease of maintenance [1]. - Existing MSRA technologies face challenges such as nonlinearity, time delays, and hysteresis, which limit their practical application [1]. Group 3: Intelligent Control Architecture - Researchers from Italy and Switzerland developed a biLSTM-based intelligent control architecture to solve the multi-functional collaborative control problem of MSRA, significantly enhancing its flexibility and situational awareness [2][22]. - The proposed method allows MSRA to perform precise operations in complex environments, such as in minimally invasive surgeries and narrow space explorations [2]. Group 4: Experimental Validation - The research team designed a cable-driven MSRA consisting of three independent modules, each approximately 0.2 meters long, driven by motors and connected by cables to ensure module independence [8][9]. - The system employs optical tracking for real-time, high-precision capture of the robot's position, providing a critical hardware foundation for training and validating the neural network [11]. Group 5: Performance Advantages - Experimental results show that the proposed method outperforms traditional methods in trajectory tracking tasks, with significantly lower tracking errors [17]. - The method also demonstrates the ability to handle complex tasks, such as maintaining specific positions while moving other parts of the robotic arm [17]. - The system successfully achieved obstacle avoidance and target tracking in dynamic environments, showcasing its adaptability [19].