Energy Consumption Analysis And Adaptive Gait Design Method For Quadruped Robots

The non-continuous nature of legged robots allows them to adapt to a wide range of terrain conditions,which has led to a great deal of research and development.Most of the tasks performed by quadruped robots require long periods of continuous operation,such as disaster relief and patrolling sites.Therefore,it is particularly essential to increase the endurance.Optimisation of gait parameters and gait transitioning at different speeds is an effective way to reduce the energy consumption of quadruped robots.The main research and innovation points of this paper are as follows:Firstly,energy consumption evaluation indicators are proposed,and energy consumption analysis of trotting is realised based on the indicators.By simplifying the physical model,the kinematics and dynamics model of the quadruped robot is constructed.Based on the kinematic characteristics of the quadruped robot,a unified energy consumption evaluation metric is also given.The trajectory planning in the swing and support phases is determined.The energy consumption corresponding to different gait parameters under trotting is calculated by analysing the motion of the single-legged model under ideal conditions.The dimensionless Froude number is used to represent the velocity,and the relationship between the gait parameters and the Froude number when the energy consumption is minimal is analysed.Conclusions with generality are obtained,which effectively guide the selection of gait parameters under trotting.Secondly,an foot location planning method based Model Predictive Control is proposed.Gait transitioning using this method between trotting and flight trotting is accomplished.Based on the numerical analysis of the dynamic model,the gait parameters that reflect the characteristics of the two gaits are selected for trotting and flight trotting,respectively.The intersection point’s velocity of the two curves is the moment for gait transitioning.To reduce the postural oscillation in gait transitioning,a foot location planning method is proposed based on MPC.By considering the quadruped robot’s body speed,posture and planned foot-end forces,the foot location of the swing leg is optimised to effectively reduce postural oscillations after gait transitioning.Thirdly,adaptive optimisation of gait parameters for quadruped robots is accomplished using reinforcement learning.The energy consumption for different gait parameters is calculated by interacting with the simulation environment.The gait period and the height of the quadruped robot are discretized and DDQN is used to learn the energy consumption under different gait parameters.Based on the results of the numerical analysis and the simulation results,the search range of the gait parameters is reasonably determined to improve the learning efficiency of reinforcement learning.In order to verify the method,the validation of the proposed method was completed using the simulation platform of MIT.A comparison of the posture after gait transitioning between the heuristic foot location planning method and the MPC-based foot location planning method is made in the simulation environment,effectively demonstrating the superiority of the proposed method.The co-compilation of the two programming languages is accomplished by establishing a communication mechanism between the different programming languages.The optimization of energy consumption-based gait parameters using DDQN is completed in a unified framework adopting pytorch.The simulation experimental results prove that the proposed gait parameter optimisation method can reduce the energy consumption of the quadruped robot.