Yue Tianqi, Lu Chenghua, Tang Kailuan, Qi Qiukai, Lu Zhenyu, Lee Loong Yi, Bloomfield-Gadȇlha Hermes, Rossiter Jonathan
School of Engineering Mathematics and Technology, University of Bristol, Bristol, UK.
School of Automation and Intelligent Manufacturing, Southern University of Science and Technology, Shenzhen, China.
Sci Robot. 2025 May 14;10(102):eadr4264. doi: 10.1126/scirobotics.adr4264.
Octopuses exploit an efficient neuromuscular hierarchy to achieve complex dexterous body manipulation, integrating sensor-rich suckers, in-arm embodied computation, and centralized higher-level reasoning. Here, we take inspiration from the hierarchical intelligence of the octopus and demonstrate how, by exploiting the fluidic energy and information capacity of simple suction cups, soft computational elements, and soft actuators, we can mimic key aspects of the neuromuscular structure of the octopus in soft robotic systems. The presented suction intelligence works at two levels: By coupling suction flow with local fluidic circuitry, soft robots can achieve octopus-like low-level embodied intelligence, including gently grasping delicate objects, adaptive curling, and encapsulating objects of unknown geometries, and by decoding the pressure response from a suction cup, robots can achieve multimodal high-level perception, including contact detection, classification of an environmental medium and surface roughness, and prediction of an interactive pulling force. As in octopuses, suction intelligence executes most of its computation within lower-level local fluidic circuitries, and minimum information is transmitted to the high-level decision-making of the "brain." This development provides insights into octopus-inspired machine intelligence through low-cost, simple, and easy-to-integrate methods. The presented suction intelligence can be readily integrated into fluidic-driven soft robots to enhance their intelligence and reduce their computational requirement and can be applied widely, from industrial object handling and robotic manufacturing to robot-assisted harvesting and interventional health care.
章鱼利用高效的神经肌肉层级结构来实现复杂灵巧的身体操控，整合了富含传感器的吸盘、手臂内的具身计算以及集中式的高级推理。在此，我们从章鱼的层级智能中汲取灵感，并展示了如何通过利用简单吸盘、软计算元件和软致动器的流体能量和信息容量，在软体机器人系统中模仿章鱼神经肌肉结构的关键方面。所呈现的吸盘智能在两个层面上发挥作用：通过将吸流与局部流体回路相耦合，软体机器人能够实现类似章鱼的低级具身智能，包括轻柔抓取 delicate 物体、自适应卷曲以及包裹未知几何形状的物体，并且通过解码来自吸盘的压力响应，机器人能够实现多模态高级感知，包括接触检测、环境介质和表面粗糙度的分类以及交互式拉力的预测。如同在章鱼中一样，吸盘智能在较低级别的局部流体回路中执行其大部分计算，并且最少的信息被传输到“大脑”的高级决策中。这一进展通过低成本、简单且易于集成的方法为受章鱼启发的机器智能提供了见解。所呈现的吸盘智能能够很容易地集成到流体驱动的软体机器人中，以增强它们的智能并降低其计算需求，并且可以广泛应用，从工业物体处理和机器人制造到机器人辅助收获和介入医疗保健。
