Developing both robust and agile locomotion skills for legged robots is non-trivial. In this work, we present the first blind locomotion system capable of traversing challenging terrains robustly while moving rapidly over natural terrains. Our approach incorporates the Adversarial Motion Priors (AMP) in locomotion policy training and demonstrates zero-shot generalization from the motion dataset on flat terrains to challenging terrains in the real world. We show this result on a quadruped robot Go1 using only proprioceptive sensors consisting of the IMU and joint encoders. Experiments on the Go1 demonstrate the robust and natural motion generated by the proposed method for traversing challenging terrains while moving rapidly over natural terrains.

More and more state-of-the-art robots have employed hydraulic actuating systems. It has a high power-to-weight ratio. Robots with these actuators can bear more payloads and achieve highly dynamic performance. However, the energy consumption is also very high and the system is very complicated comparing to the electronic motor actuated robot. A lot of research has been done to save the energy. Among which the application of springs is one of the most commonly used methods. This paper presents another use of the spring to save the energy by reducing the hydraulic system pressure of a newly built robot called the “Baby Elephant.” The configuration of the spring is designed according to the leg mechanism. The spring gives an assist force in the stance phase of the leg and exerts a passive payload in the swing phase. The maximum cylinder force is then reduced so as to bring down the pump pressure. The energy to be saved depends on how much the hydraulic pressure can be reduced. In this paper, the Baby Elephant is briefly introduced, the design of the springs on saving the energy are described. Simulations and experiments are carried out to confirm the effect.

Fault tolerance is essential for quadruped robots when they work in remote areas or hazardous environments. Many fault-tolerant gaits planning method proposed in the past decade constrained more degrees of freedom(DOFs) of a robot than necessary. Thus a novel method to realize the fault-tolerant walking is proposed. The mobility of the robot is analyzed first by using the screw theory. The result shows that the translation of the center of body(CoB) can be kept with one faulty actuator if the rotations of the body are controlled. Thus the DOFs of the robot body are divided into two parts: the translation of the CoB and the rotation of the body. The kinematic model of the whole robot is built, the algorithm is developed to actively control the body orientations at the velocity level so that the planned CoB trajectory can be realized in spite of the constraint of the faulty actuator. This gait has a similar generation sequence with the normal gait and can be applied to the robot at any position. Simulations and experiments of the fault-tolerant gait with one faulty actuator are carried out. The CoB errors and the body rotation angles are measured. Comparing to the traditional fault-tolerant gait they can be reduced by at least 50%. A fault-tolerant gait planning algorithm is presented, which not only realizes the walking of a quadruped robot with a faulty actuator, but also efficiently improves the walking performances by taking full advantage of the remaining operational actuators according to the results of the simulations and experiments.

The piezoelectric actuator exhibits both linear dynamics of creep and vibration and nonlinear dynamics of hysteresis. The model identification from experiments can compensate uncertainties in the model and get a satisfactory model for the control. Block-oriented nonlinear model is suitable for modeling this kind of linear and nonlinear hybrid dynamics. In this study, a Wiener model identification method for the piezoelectric actuator is proposed to model the linear and nonlinear dynamics together from the data. The model is called two-linear(linear+linear)-nonlinear (TLN) Wiener model, where two additive linear dynamic models are used to identify the creep and vibration dynamics, and a nonlinear dynamic model is followed to identify the hysteresis nonlinear features. The TLN-Wiener model is parameterized using Laguerre functions and Prandtl-Ishlinskii (PI) model, and the model identification algorithm is proposed. The experiments on piezoelectric actuators show that the TLN-Wiener model identification method can model the creep, vibration and hysteresis dynamics. The method is applicable to model the rate-dependency dynamics of the piezoelectric actuator.