Accurate, precise and robust motion control of dynamic quadrupedal walking robots.

动态四足步行机器人有许多潜在的应用，因为它们的腿使它们相对于车轮实现了更好的机动性和敏捷性。这些自主机器有能力执行各种被认为对人类来说太危险或无法完成的任务。他们可以被派去评估受灾地点、抢救生命、处置危险物品、搬运笨重的货物或进行检查--在不利的环境或具有挑战性的条件下。这项功能具有重要的社会价值，这项技术的成熟版本将被适用的大型私人和公共组织获得。世界上其他领先的大学、机构和公司对四条腿机器人产生了浓厚的兴趣。IIT与Moog Inc.合作制造了HyQ2Max机器人。从麻省理工学院剥离出来的波士顿动力公司构建了几个类似的产品，包括BigDog和SpotMini。Anybotics是从ETH剥离出来的，开发了ANYmal；牛津大学的Dynamic Robot Systems小组可以获得这种机器人的一个版本，这将是这项研究项目的重点。这台多功能机器的腿由安装在关节上的12个电动马达驱动，以促进全身控制技术，使动态操作，如跳跃或奔跑。机器人的运动学结构旨在实现大范围的移动性，使其能够克服障碍和楼梯。大量的研究集中在周期性步态上。然而，一个真正动态的机器人需要能够自主和实时地穿越复杂的拓扑结构，并拒绝不可预见的外部脉冲负载扰动。该项目的目标是开发低级运动控制算法，该算法将智能地僵硬和放松机器人的腿，以有效地稳定其站姿，或者允许它在保持平衡的同时执行苛刻的运动计划。我们将探索简单的串级PID控制，并依赖于扭矩测量。使用全状态反馈可能会潜在地提高关节的可控性。在自抗扰控制的基础上增加一个非线性扩张状态观测器来补偿组合扰动，将使机器人本身具有更强的鲁棒性。虽然这些方法在同行评议的文献中得到了广泛的讨论，但将它们应用于一个真正的、复杂的、有12个电机的四足机器人系统是一个原创性的来源。2018年，Winkler等人。建议通过轨迹优化生成可行的运动计划，而不强制实施固定的步态序列或立足点位置。取而代之的是，非线性求解器迭代地确定哪组参数和决策将产生成功的运动过程。虽然这一主要的理论想法很有希望，但在实验中并没有得到充分的检验。它还依赖于对物理模型的某些假设和简化，这些假设和简化偏离了现实世界。建立在这种方法的基础上，将产生一种新的动态运动方法。该项目属于EPSRC工程研究领域。这项研究由Anybotics和Moog Inc.支持。参考：A Winkler，D belicoso，M Hutter，J Buchli。基于相位化末端效应器的腿部系统步态和轨迹优化。IEEE机器人和自动化通讯(RA-L)3,1560-1567

Dynamic quadrupedal walking robots have many potential applications because their legs let them achieve superior mobility and agility relative to wheels. These autonomous machines have the capability to perform various tasks that are deemed too dangerous for or unachievable by humans. They can be sent to assess disaster sites, rescue a life, dispose of dangerous goods, carry unwieldy loads or carry out an inspection-in unfavourable environments or challenging conditions. This functionality has a significant social value and the mature version of this technology will be acquired by applicable large-scale private and public organisations.Four-legged robots have seen a major interest from other world-leading universities, institutions and companies. The IIT in a partnership with Moog Inc. built the HyQ2Max robot. Boston Dynamics, a spin-off from the MIT, constructed several similar products including BigDog and SpotMini. ANYbotics, a spin-off from the ETH, developed ANYmal; a version of this robot is available to the Dynamic Robot Systems group at the University of Oxford and will be the focus of this research project.The legs of this versatile machine are driven by twelve electric motors mounted at the joints to facilitate whole-body control techniques enabling dynamic manoeuvres such as jumping or running. The kinematic structure of the robot is designed to achieve large mobility allowing it to overcome obstacles and stairs. Significant amount of research is focused on periodic gaits. However, a truly dynamic robot needs to be able to traverse complex topologies and reject unforeseen external impulsive load disturbances autonomously and in real-time. The aim of the project is to develop low-level motion control algorithms which will intelligently stiffen and relax the legs of the robot to either stabilise its stance effectively or allow it to execute demanding motion plans while maintaining balance.We will explore simple, cascaded PID control and rely on torque measurements. The use of a full-state feedback could potentially improve the controllability of the joints. The addition of a nonlinear extended state observer, which is derived from active disturbance rejection control, to compensate for the combined disturbances would make the robot inherently more robust. While these approaches have been discussed extensively in peer-reviewed literature, applying them to a real, sophisticated, four-legged robotic system with twelve motors is a source of originality.In 2018, Winkler et al. proposed to generate feasible motion plans via trajectory optimisation without imposing a fixed gait sequence or foothold location. Instead a nonlinear solver iteratively determines what set of parameters and decisions will yield a successful course of motion. While promising, this mainly theoretical idea was not sufficiently tested in experiment. It also relies on certain assumptions and simplifications of the physical model which deviate from the real world. Building upon this method would result in a novel approach to dynamic motion. This project falls within the EPSRC Engineering research area. The research is supported by ANYbotics and Moog Inc.Reference:A Winkler, D Bellicoso, M Hutter, J Buchli. Gait and Trajectory Optimization for Legged Systems through Phase-based End-Effector Parameterization. IEEE Robotics and Automation Letters (RA-L) 3, 1560-1567