Abstract:
The evolution of robotic control strategies has progressed from velocity-level paradigms to acceleration-layer approaches, driven by the need for improved adaptability, precise force modulation, and efficient real-time execution. This article presents the acceleration layer continuous quad neural dynamics (ACQN) control method, designed to enhance adaptability and precision in both continuum and rigid robots. ACQN provides direct acceleration-layer control by advancing beyond velocity-based control, ensuring smoother operation and greater robustness. To enable implementation in digital environments, ACQN is discretized using two methods: the Verlet method for stability and time-reversibility and a modified implicit left-and-right 3-step (ILR3S) formula with simplified derivative computation using the Euler backward approach to form acceleration layer discrete quad neural dynamics (ADQN) method. Simulations in MATLAB and CoppeliaSim, alongside physical experiments, validate the proposed ADQN method, demonstrating its effectiveness in maintaining joint properties, improving control performance, and extending the robotic manipulator lifespan.
在对提升适应性、实现精确力调节及高效实时执行的需求驱动下，机器人控制策略已从速度层范式演进至加速度层方法。本文提出了一种加速度层连续四元神经动力学（ACQN）控制方法，旨在提升连续体机器人与刚性机器人的适应性与控制精度。ACQN 突破了基于速度的控制局限，实现了直接的加速度层控制，从而确保了机器人运行的平顺性与鲁棒性。为适应数字环境下的实现需求，ACQN 采用了两种离散化方法：一是利用 Verlet 方法以确保系统的稳定性和时间可逆性；二是采用一种改进的隐式左右三步（ILR3S）公式，该公式通过欧拉后向法简化了导数计算过程，从而构建了加速度层离散四元神经动力学（ADQN）方法。通过在 MATLAB 和 CoppeliaSim 平台进行的仿真实验，结合实际物理实验验证，本文所提出的 ADQN 方法得到了充分证实；实验结果表明，该方法在保持关节特性、提升控制性能以及延长机械臂使用寿命方面均展现出了显著的成效。