Study On Rhythmic Motion Control Framework For Small-Sized Quadruped Robots

Compared with wilderness exploration and rugged terrain transportation,indoor and urban services have become main application scenarios for quadruped robots.For these scenarios,quad-ruped robots with small or medium size result in higher operability and agility.However,due to the constraint space of small-size quadrupeds,high-performance sensors or actuators cannot be equipped,while the multi-task control objectives in indoor and urban service scenarios still need to fulfill.Thus,a new type of control framework is expected to meet the requirements of simplicity and effectiveness,so that a small or medium-sized quadruped robot could be ad-equately controlled to achieve different objectives in indoor and urban service scenarios.For con-ventional model-based hierarchical optimal control,it can achieve excellent dynamic performance but usually requires precise dynamic modeling and high-performance torque-controlled actuators.Biologically-inspired control has a simpler architecture but relies on experience-based parameter tuning processes,where a good control performance cannot always be guaranteed.Thus,these two types of control framework cannot suitably solve the problem of locomotion control for small or medium-sized quadruped robots under indoor and urban service scenarios.In order to address the above problem,we proposed a biologically inspired multi-module framework for quadrupedal rhythmic motion generation.This framework combines conventional model-based methods with biologically inspired methods,which results in the simplicity and ef-fectiveness of the controller.According to the biology concept of motor primitives,this framework is in parallel structure and generates complex motions by linearly superimposing basic and super-posed motions.These motions are expressed as foot-end trajectories but with different functional-ities.Basic motions form the essential moving gait by defining basic walking characteristics such as stride length,step height and step period.Superposed motions increase stability and agility,drive the robot to move laterally and rotationally.These primitive motions are generated through different parallel modules.Thanks to the central pattern generator model used in the framework,the framework can smoothly switch the robot state between forward,lateral and rotational mov-ing.No online motion replanning is required.Notably,this framework only provides the general structure of the controller,and detailed motion definitions vary according to the applied scenarios.Based on the proposed framework,we studied the locomotion control for quadruped robots in three scenarios:1.Multi-state control for quadrupedal walking on flat surfaces.In this project,we solved the fundamental control problem of start,walking and stop.Zero moment point control and foot-end trajectory compensation were used as feedforward control to realize a stable walking gait Making use of simple kinematic laws and the characteristic of the proposed framework,smooth state switch between forward,backward,lateral,and rotational moving was achieved.Moreover,push adaptation control under walk gait was briefly discussed in a simulation.2.Human-robot physical interaction based on lateral pushes.In this project,we first realized the trotting gait and the lateral moving through feedforward foot-end trajectory compensation.Secondly,taking advantage of the spine joint that the robot has,we proposed a control approach for stabilizing the robot roll angle through virtual spring control and stability margin control.This approach allowed the robot to maintain balance under strong lateral pushes while staying on its original moving course.Finally,a new pattern of human-robot physical interaction was achieved,where small lateral pushes made the robot move laterally on its own initiative,while strong lateral pushes made the robot continue moving on its original course.3.Untethered quadrupedal jumping control on a trampoline.In this project,we realized quadrupedal pronking/hopping on a trampoline while performing translational and rotational mo-tions.The multi-module framework was used in this project to combine foot-contact-forces based balance keeping motions with kinematic-laws based translational or rotational motions.We also analyze the convergence of the pronking height,and briefly illustrated the differences and simi-larities between jumping on compliant surfaces with stiff legs and jumping on hard surfaces with springy legs.Thus,this project could provide an algorithmic design reference for the subsequent implementation of leg elastic mechanism design and hard surface jumping.All the controller designs in the above projects were verified on the quadruped robot Horizon designed by Zhejiang University Robotic Lab or the compliant quadruped robot Serval designed by EPFL Biorobotics Lab,which proved the simplicity,effectiveness and universalizability of the proposed framework.