Advances in legged robots control, perception and learning

This is the Institution of Engineering and Technology (IET) Cyber-systems Robotics Special Issue Advances in Legged Robots: Control, Perception Learning. mammals are found everywhere nature. These legged animals can reach anywhere on Earth, adapt to uneven, discontinuous obstructed environment. Their locomotion patterns flexible diverse better living Imitating these real animals, robots have potential advantages over wheeled or tracked vehicles regard traversal rough unstructured terrain. However, there still many challenges for systems solving technical problems world. perception learning key technologies field robots. Control basis stable robot. The combination control mechatronics machines will show excellent passing ability continuous stairs, discrete terrain vertical obstacle environments. methods mainly include model-based learning-based control. After decades development, research results made more stable. latter a new method combining artificial intelligence, exploring how acquire motor skills process interacting with environment achieve expected abilities. allows robot understand Autonomous navigation, behavioural decision-making task operation, all require environmental awareness understanding. an unattainable component For example, different road surfaces gait modes, which most direct perceptual demand List: Paper 1: Disturbance rejection biped during walking running using moment gyroscopes, 2: A self-stabilised humanoid based essential model internal states, 3: Squat motion bipedal real-time kinematic prediction (RKP) whole-body control, 4: Design robotic system legs, wheels reconfigurable arm, 5: heuristic framework heavy-duty hexapod complex terrain, 6: Efficient robust quadruped bounding pretrained neural networks, 7: pneumatic muscle (PM) bionic network, 8: Sampling visual simultaneous localization mapping (SLAM) wide-angle camera mobile robots, 9: Research autonomous manipulator realize leader-following, object recognition, navigation operation. 1 by Haochen Xu, paper 2 Qiuyue Luo 3 Wenhan Cai investigated 4 Linqi Ye studied leg–arm wheel reconfiguration design strategy 5 Jinmian Hou extended multi-leg problems. design, discussed. In 1, Xu et al. disturbance CMG. They used CMG as auxiliary stabilisation device fully actuated integrated into balance framework. effectively help resist remain time. 2, exploited states. dynamic full zero dynamics concept. By adjusting step timing landing position swing foot automatically following intrinsic characteristics, walking. 3, proposed squat RKP considers upcoming reference trajectories combines it quadratic programming (QP)-based whole body (WBC). It greatly reduces computational cost compared predictive WBC exhibits high adaptability planning much less computation 4, designed takes advantage legs. this system, walk, reducing total weight robot, multi-task was described variable configuration complicate 5, introduced novel traverse torque controller, virtual optimise ground reaction forces tracking pre-planned single rigid-body dynamics. 6 Zhicheng Wang deep reinforcement (DRL) problem, while 7 Chaoyue network problem. Both aspects DRL approach learn variety gaits efficiently adaptation. improve performance non-linear actuators. 6, efficient networks. their method, they reward function enforce symmetry periodicity learnt feedback controller simulation where build environments learning. has been deployed 7, PM network. double closed-loop leg designed. Based three-element model, feedforward neuron proportion-integral-derivative inner loop, sliding mode local approximation radial outer loop. 8 Guangyu Fan SLAM problem environments, 9 Jiamin Guo recognition perception. robustness localization, enabling tasks Additionally, environment, make decisions move destinations. 8, sampling sampled image blocks clear texture enhanced details extract feature points, matched points were extracted from images. selected template relationship between images established through pixel coordinates descriptors calculated. 9, researched realise paper, authors developed systematic solution including hardware software architecture. operation manipulator, be other forms All papers special issue indicate trends emerging We hope that benefit researchers around world exchanging sharing latest results. guest editors would like thank IET editors-in-chief Rob Buckingham, Jian Chu Max Meng continued support. Finally, acknowledge peer experts who contributed provided helpful reviews. Qiuguo Zhu received his bachelor’s degree Mechanical College Zhejiang University master’s Science 2008 2011, respectively. Then, he Ph.D. 2020. His interests quadrupedal running, manipulators machine intelligence. As PI, obtained funds organizations such National & Natural Foundation, associate editor Cyber Systems Robotics. Rui Song Shandong Technology, Jinan, China 2001, doctoral theory engineering University, 2011. He currently professor supervisor School at University. also working Vice President Institute Technology. Deputy Head Expert Group Ministry Technology's “14th Five-Year Plan” Intelligent Key Project Guidelines. Professor awarded distinguished Cheung Kong Scholars Program. won second prize third Progress Award Province. During past five years, granted than 40 national invention patents published 50 SCI/EI papers. Jun Wu B.Eng Ph.D degrees both industrial automation Huazhong China, 1989 1994, Since faculty member Engineering, professor. Dr senior IEEE Royal Society K.C. Wong Fellowship. 80 journals conference proceedings 1997/1998 IEE Heaviside Premium. Yamakita Masaki BE, ME PhD Tokyo 1984, 1986 1989, From department institute 1993 lecturer Toyohashi mechanical robotics, IEEE, SICE RSJ. Zhangguo Yu director Institute, Beijing China. B.S. electronics M.S. Southwest (SWUST), 1997 2005, respectively, (BIT), Beijing, 2009. 2000, Engineer Changhong Electric Company Ltd., 2001 2009, staff SWUST. visiting scholar Waseda Tokyo, Japan, 2008. served Program Chair ARSO 2019, Co-Chair Humanoids 2018, ICMA 2015 International Journal Humanoid current