Research On Design And Predictive Control Of Human Following Rehabilitation Robot

In order to promote the health of the elderly and the disabled and to fill the vacancy of professional nursing staff,it is urgent to develop rehabilitation robots.In the current study,rehabilitation robots such as intelligent walking aids and smart canes can provide users with services such as auxiliary support,navigation and fall protection.However,since most of these robots move in front of people close to them,so users are prone to be blocked and choose to walk in small strides,which will affect the experience when walking and the effect of rehabilitation.Inspired by the fact that rehabilitation physicians follow patients side-by-side and provide timely assistance,we tried to develop a Human Following Rehabilitation Robot(HFRR)that can follow the user side-by-side and provide auxiliary support when needed,which can be used to assist patients with lower limb muscle weakness to carry out basic walking rehabilitation training.Firstly,a Human Following Rehabilitation Robot was designed,which is composed of a differential mobile base,sensors and an embedded computer Jetson TX2.The HFRR detects the 3D coordinates of key points of the human body through Open Pose and Kinect,detects the user’s orientation through an IMU sensor,and records its own position and orientation according to the odometer.These functional nodes are synchronized and communicated based on the robot software system.Then,a walking trajectory prediction algorithm based on the Seq2 Seq model was proposed,which predicts the future walking trajectory sequence based on the historical walking state sequence.Aiming at the problem of small amplitude periodic swings of the orientation and key point trajectories when walking,a multi-modal data fusion algorithm based on Kalman filter was proposed to obtain relatively smooth walking state sequences.Furthermore,the dataset of walking trajectory was collected by the HFRR,and the experiments verified that the proposed walking trajectory prediction algorithm can accurately predict the future walking trajectory of the target in a short period of time.Finally,a control algorithm for following side-by-side based on Model Predictive Control(MPC)and walking trajectory prediction was proposed.After designing an error model,predictive model,objective function,and constraints,the problem of controlling the robot following side-by-side was transformed into the problem of solving quadratic programming in each control interval.Through simulation experiments,the effects of MPC and Lyapunov’s direct method for controlling the robot following side-by-side were compared.The comparison results show that the overall control performance of MPC is better than that of Lyapunov’s direct method based control(LDMC)because MPC comprehensively considers the constraints of control variables,the penalty items of control increments and the prediction trajectory of targets.In addition,the effectiveness of the proposed control algorithm was verified by two experiments of straight walking and turning with the HFRR.