Search papers, labs, and topics across Lattice.
This paper addresses the challenges of load transportation with quadruped robots by introducing an impedance-aware Zero Moment Point (ZMP) model that incorporates passive payload-interface dynamics. By analyzing the effects of stiffness, damping, and payload mass on stability margins, the authors demonstrate that underdamped configurations can resonate with locomotion harmonics, leading to significant improvements in stability and efficiency. The proposed Model Predictive Control framework reduces stability violations by up to 10 times and enhances locomotion efficiency by decreasing horizontal ground reaction forces by up to 15%, validated through both simulations and hardware experiments.
Stability violations in quadruped robots can be reduced by up to 10x through impedance-aware control strategies that leverage passive dynamics.
Load transportation with quadruped robots is strongly affected by the dynamics of the physical interface between the robot and the load. Passive spring-based arms reduce weight and complexity compared to active manipulators, but their spring-damper dynamics can introduce oscillatory forces that degrade locomotion stability. This paper derives an extended Zero Moment Point (ZMP) formulation that includes passive payload-interface dynamics, relating stiffness, damping, and payload mass to the stability margin. The analysis shows that underdamped configurations can resonate with locomotion harmonics. Based on this insight, we augment a Single Rigid Body Dynamics model with passive subsystem dynamics and integrate it into a Model Predictive Control framework. In simulation, the proposed controller reduces stability violations by up to $10\times$, from $7.0\%$ to $0.7\%$, and increase locomotion efficiency by lowering horizontal ground reaction force effort by up to $15\%$ compared to a nominal baseline. Hardware experiments with a $2\,\mathrm{kg}$ payload show stable locomotion under pull-release disturbances where the nominal controller fails. The same model also enables end-effector tracking through passive arm dynamics without direct arm actuation.
