Balance maintenance of a humanoid robot using the hip-ankle strategy

Abstract

Prevention of falling is of major importance in the practical application of humanoid robots. This thesis focuses on the hip-ankle strategy as a method to maintain balance when subjected to a large disturbance. This strategy is characterised by a large rotation of the hip joint which repositions the Centre of Mass (CoM). The hip-ankle strategy is compared to the ankle strategy which locks the hip joint and compensates for a disturbance by applying ankle torque. Various hip-ankle strategy controllers have been presented in literature of which three implementations on humanoid robots are known. However, none of the research provided experimental evidence that a humanoid robot can withstand larger disturbances by using the hip-ankle strategy compared to the ankle strategy. The goal of this research was therefore to provide experimental evidence that a humanoid robot can maintain balance for larger distur- bances by using the hip-ankle strategy than solely using the ankle strategy. First, simple models were used to investigate the theoretical maximum allowable disturbance for the humanoid robot TUlip. The Inverted Pendulum Model (IPM) was used to simulate the ankle strategy where ankle torque was the only control input. The hip-ankle strategy was simulated using the Inverted Pendulum plus Flywheel Model (IPFM) where flywheel torque served as an additional control input. The hip-ankle strategy was implemented by using bang-bang control input profile on the flywheel. Disturbances were applied in horizontal direction to the CoM of the model and measured in terms of applied impulse. In these simulations the hip-ankle strategy was able to maintain balance for disturbances 33.6% larger than the ankle strategy. Next, a control algorithm was developed for the humanoid robot TUlip. In the control algorithm, the hip-ankle strategy was implemented by the application of virtual forces and torques on the trunk of the robot. These virtual forces were then transfered to joint torques by means of Virtual Model Control (VMC). The Instantaneous Capture Point (ICP) of the robot was controlled to a desired location by modulating the Centre of Pressure (CoP) of the robot. If the CoP is sufficient to keep the ICP within the foot, the control algorithm will lock the hip joint and balance using solely the ankle torque. In case the ICP crosses the edge of the foot, a large virtual torque on the upper body is applied in the direction of the ICP which results in a rotational acceleration of the upper body. The ICP will then be pushed back into the foot and the application of ankle torque will then be sufficient again to maintain balance. In order to evaluate if it is physically possible for the humanoid robot TUlip to maintain balance using the hip-ankle strategy control algorithm, a simulation was performed. The Double Inverted Pendulum Model (DIPM) was used as a model for TUlip and constraints in terms of maximum joint torque and range of motion were included. The ankle strategy and the hip-ankle strategy were then subjected to horizontal impulsive disturbances to the back of the upper body of the robot and measured in terms of impulse. The hip-ankle strategy was able to withstand 16.6% larger pushes than the ankle strategy. Finally, the hip-ankle strategy was experimentally evaluated on the humanoid robot TUlip. Disturbances were created by swinging a weight at the end of a pendulum to the back of the upper body of the robot. The applied impulse was measured by measuring the disturbance force over time by using a load cell. Experimental results showed that the hip-ankle strategy implemented on the humanoid robot can maintain balance for disturbances 18% larger than solely using the ankle strategy. The main conclusion of this thesis is that the hip-ankle strategy can be used to improve the balance maintenance on a humanoid robot. This thesis provided the first experimental evidence that a humanoid robot can maintain balance for larger disturbances by using the hip-ankle strategy than using the ankle strategy.