Body State Estimation Of Legged Robots Based On Multi-sensor Information Fusion

Legged robot is an important application field of artificial intelligence,which has attracted more and more attention from various researchers in recent years.As feedback information in control,it is very important to accurately estimate the body velocity and position,which can promote the robot's performance.Information from the Inertial Measurement Unit(IMU)or kinematics analysis both can be used to estimate the body state,but drift errors of the IMU estimation and noise errors of the kinematics estimation cause that both methods are unable to meet the realtime control demand.Therefore,a body state estimation method based on the information fusion between IMU and kinematics was proposed in this paper.The main research work is as follows:(1)Body velocity and position was defined as the state variables in this paper according to the research demand.And to solve the body state,IMU information was processed by the velocity updating differential equations and related coordinate frames were transformed.On account of the drift error of the IMU information,it was calibrated and compensated,in which a Six-Position approach was employed.(2)Kinematical models of the single leg and the whole quadruped robot were built based on the sensors information and geometrical configuration.And by analyzing the model,location of feet points relate to the body center frame were calculated.On the contrary,the state of body center point relate to the feet point of the supporting leg can be obtained.(3)A multi-sensor information fusion strategy was proposed to estimate the body state.First level fusion was conducted among the supporting legs information,where the quartz force sensor played an important role in judging legs' supporting state.After that,Kalman theory was employed to fuse the IMU information and the first results as the second level fusion.Then the optimal estimation of the body state can be gained.Simulations in RecurDyn/Matlab was conducted and its results verified the availability of the fusion method in estimating the forward and vertical state of the body.(4)Motion Analysis System was built to measure the body velocity and position,which would be compared to the fusion estimation value as the truth-value.Experiments in single leg and the quadruped robot were designed to verify the method's effectiveness.And results show that compared with the estimation based on single sensor,multi-sensor information fusion method can get higher accuracy in estimating the body state.