Trajectory generation for flight phase of a quadruped robot jump

Official URL: https://risc01.sabanciuniv.edu/record=b2760676_(Table of contents)

Legged robots excel in navigating challenging natural environments, such as steep obstructions or wide gaps in the ground. Apart from rough terrain, they may confront unexpected impact forces during their leaping gaits. While facing external disturbances, legged robots should maintain and restore their stability while completing their gaits. External disturbances and body orientation errors should be identified. Appropriate actions have to be taken to restore the balance of the robot and provide advantageous landing circumstances. This dissertation examines the robot body orientation errors during the flight phase and first offers a unique posture control method that uses reinforcement learning to build reference trajectories for a quadrupedal robot with waist joints during a long jump flight phase. Then, another novel algorithm for posture recovery is provided, this time based on angular momentum. The same algorithm is altered to account for perturbations in the flying phase caused by a push on the robot’s body. We describe a push recovery method that uses angular momentum to build reference trajectories for the long jump. This work also contains a more detailed angular momentum-based reference generating approach for posture recovery. Real-time centroidal dynamics computation is employed in this second technique. These approaches provide reference trajectories for the waist and rear hip joints of the quadrupedal robot in order to acquire the desired orientation of the robot in the air. PID joint position control is used to track reference trajectories. The robot model used in the calculations is comprehensive since each component of the robot’s body—the leg links and three torso portions—is represented by individual parameters. The suggested techniques for trajectory creation are computationally efficient, making them suited for use in real-time applications. The proposed posture control and push recovery approaches are tested on the model of a quadrupedal robot during the flight phase of a long jump via simulations. The findings reveal that the proposed methods are accurate in terms of angular position and angular velocity regulation and can achieve successful landing postures.