Bipedal Robot Motion Control Based On Hydraulic Series Elastic Actuator

The bipedal robots have become the new frontier in robotic research due to their superior adaptability to complex terrains such as ravine,abrupt slope,and steps in comparison to the wheeled or tracked robots.With the increasing need for compliant actuation and high mobility,the series elastic actuator(SEA)has been considered as a promising device for achieving bipedal walking owing to the advantages of low output impedance,impact absorption,and energy storage.However,traditional SEA-based control methods have several disadvantages that can severely limit the robot's performance.For actuation control,current compliant control methods can introduce large tracking errors and time delays.Moreover,approaches for hydraulic SEA control are not well investigated.On the other hand,for gait generation,current methods cannot meet the requirements for high walking speed,good terrain adaptability,and online generation.To cope with these problems,this thesis aims to make contributions to improve the robot performance in terms of both hydraulic SEA control and gait generation.The proposed methods are further applied to realize fast bipedal walking on uneven terrain in simulation.The main contributions are as follows.Firstly,a high-precision admittance control method is proposed for the hydraulic SEA.Owing to the non-backdrivability of the hydraulic actuation,the hydraulic cylinder movement can be considered as a position source,in which the tracking error is handled as the input disturbance,thereby simplifying the system model.An admittance controller that combines a passive disturbance observer(P-DOB)and a load position feedback compensator is developed to reduce the tracking error.The control law of the proposed framework decouples the complicated hydraulic dynamics and therefore facilitates the practical application for hydraulic SEA and enhances the tracking accuracy.The experimental results verify the effectiveness of the proposed method.Secondly,a delay compensation method is proposed for the hydraulic SEA.A lowdelay admittance control scheme is developed,in which the adaptive time series(ATS)compensator is adopted to compensate for the time delays and errors in both load movement and external force estimation,thereby improving the overall performance.Moreover,a parallel adaptive time series(P-ATS)compensator is also proposed with an enhanced parameter determination mechanism,thereby improving its compensation ability and performance under low-sampling-rate conditions.The experimental results suggest a high-performance stiffness tracking can be achieved through the proposed method,and the P-ATS compensator outperforms the ATS compensator.Thirdly,a kinematic calibration method is proposed for compliant leg control.Based on previous studies,a stiffness control scheme is first constructed for a compliant leg actuated by the hydraulic SEA.To eliminate the tracking error introduced by the inaccurate kinematic model,an improved whale swarm algorithm(IWSA)-based calibration scheme is proposed to obtain the accurate kinematic parameters.The IWSA is a modified version of the whale swarm algorithm(WSA),in which improved initialization,boundary checking,and neighborhood searching strategies are adopted to enhance the searching ability and convergence rate.All these improvements are adopted to obtain more accurate kinematic parameters to improve control performance.The experimental results indicate a significant decrease in angle tracking error,which verifies the effectiveness of the proposal.Fourthly,an online gait generation method based on the compliant leg is proposed.To cope with the disturbances introduced by terrain height variations,an improved walking pattern removes the step length restriction and improves the maximum walking speed and terrain adaptability.Moreover,current methods are required to solve the gait generation problem by using nonlinear optimization approaches,resulting in difficulties for online control.To cope with this problem,a back propagation(BP)neural network is developed to train the gait datasets obtained by the IWSA,and then construct the online controller.The proposed method can generate high-quality gaits in a short time.The effectiveness of such method is further confirmed through the Webots simulation software.Finally,the contributions are summarized,and the future work is also discussed.