Optimization-Based Control Approaches to Humanoid Balancing

Optimal Control Combining Emulation and Imitation to Acquire Physical Assistance Skills

“…Additionally, different strategies such as hip or ankle strategies have been developed to recover the robot from unstable states [10]. Optimization-based methods have been employed to consider the locomotion of the robot while planning for the unactuated states and constraints [9], [11]. Another approach is to partition the robot into multiple kinematic chains, with some of them focusing on task performance and others being responsible for balance maintenance [12].…”

Towards postural balance control of exoskeletons

“…Last, for optimization-based controllers like the momentum-based controller the computational power can be a limitation [63]. In our experiments this was not an issue, because we used a model with only three degrees of freedom (Chapters 4 and 5).…”

“…In a momentum-based controller implementation on a humanoid robot, a simplification of the optimization process was proposed to run the controller in a fast torque control loop of 1 kHz through the application of hierarchical quadratic programming [55]. The idea of hierarchical quadratic programming is to first solve a quadratic program to obtain a solution for a high priority task, and then solve another quadratic program for a lower priority task, without decreasing the cost function of the previous task [63]. By applying this method, the computation time could be reduced with 40% compared to the case without task decomposition.…”