A Robust Quadruped Robot with Twisting Waist for Flexible Motions

Abstract

The waist plays a crucial role in the agile movement of many animals in nature. It provides the torso with additional degrees of freedom and flexibility, inspiring researchers to incorporate this biological feature into robotic structures to enhance robot locomotion. This paper presents a cost-effective and low-complexity waist mechanism integrated into the structure of the open-source robot solo8, adding a new degree of freedom (DOF) to its torso. We refer to this novel robot as solo9. Additionally, we propose a full-body control method for the waist-equipped quadruped robot based on generative adversarial imitation learning (GAIL). During training, the discriminator is used as input for iterative optimization of the policy and dataset, enabling solo9 to achieve flexible steering maneuvers across various gaits. Extensive tests of solo9's steering capabilities, terrain adaptability, and robustness are conducted in both simulation and real-world scenarios, with detailed comparisons to solo8 and solo12, demonstrating the effectiveness of the control algorithm and the advantages of the waist mechanism.

Figures (6)

• Figure 1: (a) The 9-DOF quadruped robot ‘solo9’. (b) Close-up of the waist mechanism. (c) Close-up of the waist twisted 90 degrees. (d) Front views of solo9 after folding. (e) Side views of solo9 after folding.
• Figure 2: Schematic diagram of solo9's mechanical structure: (a) (c) show the waist motor structure, (b) is the whole-body view, (d) is the top-down view.
• Figure 3: The figure illustrates the dataset-policy co-optimization control algorithm for solo9 based on GAIL. Starting with the solo8 original dataset, and after manual selection, the discriminator inputs partial observations and combines gait and constraint rewards for adaptive training of the waist, resulting in the initial solo9 dataset. Further dataset-policy co-optimization with command guidance is performed, increasing the weight of imitation rewards through iterations, ultimately producing the solo9 flexible steering dataset.
• Figure 4: The key reward functions utilized in solo9's training
• Figure 5: (a), (b), and (c) show the results of the turning task in the trot gait for solo8, solo9, and solo12 in Isaac Gym, respectively, with solo9 demonstrating the smallest and most stable turning radius. (d), (e), and (f) display the yaw rate tracking performance of solo8, solo9, and solo12 during the aforementioned turning task, respectively, with solo9 showing the most stable tracking performance.