Citation Context

...hybrid dynamics of walking robots makes multistep optimization difficult [1]. Successful examples of using MPC for humanoid control have therefore relied upon the use of low-dimensional linear models =-=[2]-=-, [3] or relaxation of constraints to permit smooth optimization through discontinuous dynamics [4]. Several researchers have recently explored using quadratic programs (QPs) to control bipedal system...

Citation Context

...ety of traditionally disparate approaches have been taken ranging from nonlinear control via input/output (IO) feedback linearization through output based control [12,34], force-based control methods =-=[7,17,30,21,31]-=-, and whole-body control methods for manipulation [14,15,27], to name a few. Each of these methods, along with the wide variety of other methods taken in the control of robotic systems, have proven su...

Citation Context

...has not enjoyed much success. Indeed, for the case of bipedal postural stability, the coordination of ankle, hip, and stepping recovery strategies has yielded impressive results on real systems (e.g. =-=[12]-=-). However, given our increasing understanding of human balance recovery, there is reason to suspect that coordination of the arms may offer significant advantages. Several researchers have studied mo...

Citation Context

...cribed analytically. These analytic relationships can be used to determine which strategies to use based on the current state of the robot. Push Recovery Model Predictive Control (See Section 4.3 and =-=[116]-=-): Model predictive control can be performed using simple models, a special objective function, and carefully chosen constraints to compute desired forces and footstep locations that account for futur...

Citation Context

...pport polygon, stable walking is guaranteed. Using ZMP preview control, high quality hardware [4,5,6] can walk reliably on flat ground. Next generation walking controllers from the ZMP preview family =-=[7,8,9]-=- also consider foot placement in addition to ZMP control either by including the footstep locations in the optimization process, or by using a simplified model to compute a footstep plan online. The a...

Citation Context

...h therefore can be seen as a focus point of all (lines of action of) leg forces. When the LIP is used for bipedal walking control, the Zero Moment Point (ZMP, [3]) is assumed to be equivalent to the LIP base joint p and is designed to be within the foot (e.g. in the foot center). In a real robot, the resulting ZMPs generally deviate from the planned focus points due to the model inaccuracies, but often not far enough to make the robot tilt and fall. This way, preplanning a feasible set of focus points (ZMP’s) is a successful method for bipedal gait generation and control: Stephens and Atkeson [4] present a Model Predictive Control (MPC) method for step adjustment and push recovery. Wieber [5] proposes a trajectory free linear MPC scheme, allowing for compensation of strong perturbations. Nishiwaki and Kagami [6] generate dynamically stable walking patterns 2013 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) November 3-7, 2013. Tokyo, Japan 978-1-4673-6357-0/13/$31.00 ©2013 IEEE 2600 by frequently updating a preview controller. Sugihara [7] introduces the Best COM-ZMP Regulator facilitating step adjustment of bipedal robots. Kajita et al. [8] demonstrate out...

Citation Context

... with constant CoM height is notably robust against terrain variations. Overall, we try to reject such disturbances by taking proper footsteps, rather than modulating ZMP and relying on ankle torques =-=[11, 3, 23]-=-. While bringing the system to the rest conditions after pushes exists in literature over a single step [23] or multiple steps [9], we aim at a more generic scenario, i.e. bringing the system to a des...

Citation Context

...ancing and Walking for Real-time 3D Characterss3sbalance when using the controller approach, as without it few other activities arespossible. Balancing in the presence of forces has been demonstrated =-=[14, 17, 18]-=- andsmore elaborate movements such as walking, running, jumping and cycling have beensachieved [8, 12, 15, 16]. Further enhancements are to augment the controller with datasdriven approaches such as m...

Citation Context

...the ground that enables the system to come to a stop if it could instantaneously place its foot there, not taking into account step time or SL constraints. When the capture region intersects the BoS, a humanoid can modulate its CoP to balance and does not need to take a step. When the capture region and BoS are disjoint, the humanoid must take a step to prevent a fall. Exact computation of the CP in a full simulation of the humanoid is difficult; the linear inverted pendulum model (LIPM) [22] or Linear Inverted Pendulum plus Flywheel model [13] are often used to approximate the biped dynamics [10,15,16,20,21,23]. Although a number of studies have dealt with algorithms for balance-recovery stepping for both human and humanoid robot, very little is known about re-initiation of forward motion following a recovery step of variable length. This motion may begin from a static position when recovery occurred by coming to a complete stop; for most of the humanoid push-recovery in our control schemes, there is a stopped or nearly stopped state [14,20,21,24,25]. Static stepping may also occur when there are pauses in gait to adapt to external constraints such as isolated foothold availability when walking on s...