Torque-Based Dynamic Walking - A Long Way from Simulation to Experiment

Balancing Control Through Post-Optimization of Contact Forces

“…After some controller tuning, we managed to have the robot balance on two feet, even when subject to small external perturbations (see Figure 4b). Despite the encouraging results, we experienced instabilities and strong vibrations, in a very similar way as described in [26]; indeed, implementation of full-torque controllers on legged robots appears to be challenging, and it needs further investigation from the authors in order to improve the robustness of the controller and successfully deploy it to real systems.…”

“…Future works will address first the implementation on the real hardware which, as stated previously, at the moment presents some stability problems, very similar to the one in [26]. Second we would like to explore a single-stage torque-force optimization, and compare it to [19].…”

Biped Stabilization by Linear Feedback of the Variable-Height Inverted Pendulum Model

“…2) Equality constraints: states and inputs are bound together by Equations (20) (updated by (34) to include ∆σ), (21) and (22). Let us omit d superscripts for concision:…”

“…3) Inequality constraints: limits (25) on the ZMP, (27) on λ and (28) on ω can be readily included as block matrices in G and h. DCM height limits are obtained by injecting the expression of the related spatial DCM velocity (34) into Equation 29:…”

“…Reduced model control produces a desired net contact wrench. For torque-controlled robots, this net wrench is supplied as a target to whole-body control [6], [33], [34], [35], [36], [37], and the resulting joint torques are sent to lower-level joint controllers. For position-controlled robots, an additional layer is required to regulate wrenches by admittance control.…”