Dynamic modeling and simulation of a torque-controlled spatial quadruped robot

Abstract

Evolution has shown that legged locomotion is most adequate for tasks requiring versatile movement on land, allowing animals to traverse a wide variety of environments ranging from natural terrain to artificial, man-made landscapes with great ease. By employing well-designed control schemes, this ability could be replicated for legged robots, enabling them to be used in critical situations that still pose great danger to human integrity, such as search and rescue missions, inspection of hazardous areas, and even space exploration. This work characterizes the quadruped robot and contact dynamics that will compose our in-house simulator to be used for prototyping locomotion control schemes applied to quadruped robots. The proposed simulator computes the robot dynamics using the Recursive Newton-Euler and Composite-Rigid-Body algorithms with a few modifications to make certain aspects relevant for contact detection and control more easily accessible; furthermore, a compliant contact force method alongside stick-slip friction modeled the contact dynamics. To allow the robot to move, a simple PD-independent joint controller was implemented to track a desired leg trajectory. With the same robot and controller implemented using the MuJoCo simulation software, this work evaluates the proposed simulator by comparing characteristic locomotion signals such as the trunk pose and the ground reaction forces. Results showed similar behavior for both simulators, especially with regard to the contact detection, despite the significantly different contact models. Lastly, final remarks to enhance our simulator's performance are suggested to be explored in future works.