Whole-Body Impedance Coordinative Control of Wheel-Legged Robot on Uncertain Terrain

Lei Shi, Xinghua Yu, Cheng Zhou, Wanxin Jin, Wanchao Chi, Shenghao Zhang, Dongsheng Zhang, Xiong Li, Zhengyou ZhangarXiv 2024 · DOI · arXiv

Introduction

Navigating real-world environments presents a significant challenge for humanoid robots, particularly when balancing the conflicting requirements of locomotion agility and manipulation stability. Heavy robots, such as the 80kg X-Man developed by Tencent Robotics X, face substantial inertial hurdles. Impacts from uneven terrain, such as cobblestones or slopes, can propagate through the robot's frame, jeopardizing balance and the safety of carried payloads. Traditional control strategies often struggle with these nonlinear dynamics and typically require precise, pre-mapped terrain data.

The X-Man wheel-legged robot demonstrating passive stability while carrying a bottle across uncertain terrains.

Proposed Methodology

To address these challenges, researchers proposed a bi-level whole-body control framework that integrates variable-damping impedance control with real-time terrain estimation.

Bi-Level Control Strategy

The framework separates high-level stability optimization from low-level torque generation:

  1. Outer Level: Variable Damping Impedance Control (ICC): Inspired by the principle of "suspended backpacks," this controller models the robot's upper body as a mass-spring-damper system. By applying Pontryagin’s Minimum Principle, the system implements a "bang-bang" control strategy for damping parameters. This allows the robot to dynamically adjust its "softness" to dissipate energy from ground impacts before they affect the torso.
  2. Inner Level: Whole-Body Control (WBC) with Terrain Estimation: The WBC layer utilizes Quadratic Programming (QP) to determine optimal motor torques across the robot's 29 degrees of freedom (DOF).
Overview of robot tasks and joint configurations used in the whole-body control framework.

Key Technical Innovations

  • Proprioceptive Terrain Estimation: Instead of relying on external vision sensors, the robot estimates the terrain frame (normal and tangent vectors) by analyzing wheel trajectories and contact forces. This allows for robust navigation on slopes and uneven surfaces without high-fidelity maps.
  • Model-free Friction Compensation: A dedicated layer compensates for joint friction and "stickiness" without requiring complex mathematical models of the internal mechanics, improving the precision of high-torque movements.

Experimental Evaluation

The framework was rigorously tested on the X-Man robot across various physical testbeds:

  • Complex Terrains: A path including flat ground, slopes up to 12.6 degrees (0.22 rad), and cobblestones.
  • Dynamic Adaptation: A "wave slope" road requiring continuous leg extension and contraction.
  • Manipulation Task: A water-carrying task where the robot transported a tray with a bottle of water, testing its ability to maintain a level base despite ground disturbances.

Key Insights and Results

The experiments demonstrated that the robot could maintain exceptional upper-body stability, with force fluctuations kept within 18 Newtons even on cobblestones. The X-Man successfully completed the water-carrying task, with its legs actively adjusting (up to a 35cm range) to keep the torso level. Furthermore, the model-free friction compensation effectively handled zero-speed transition points, preventing the "jerky" motion often seen in heavy robots when joints change direction.

Limitations and Future Directions

While the current framework provides robust stability, several areas for future research remain:

  • Dynamic Speed Planning: The current tests used a preset maximum velocity of 0.8 m/s. Integrating a speed planner that responds to terrain "noise" could enhance safety.
  • Active Manipulation: Future work will explore more complex, active manipulation tasks, such as opening doors, while traversing uncertain terrains.
  • Sensor Fusion: Combining proprioceptive data with visual "look-ahead" capabilities would allow the robot to preemptively adjust to upcoming terrain changes.

Summary updated

BibTeX
@misc{shi2024whole,
  author = {Lei Shi and Xinghua Yu and Cheng Zhou and Wanxin Jin and Wanchao Chi and Shenghao Zhang and Dongsheng Zhang and Xiong Li and Zhengyou Zhang},
  title = {{Whole-Body Impedance Coordinative Control of Wheel-Legged Robot on Uncertain Terrain}},
  year = {2024},
  eprint = {2411.09935},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2411.09935},
}
Figure