Convex Optimization in Legged Robots

Prathamesh Saraf,Mustafa Shaikh,Myron Phan

2023-07-01

Abstract:Convex optimization is crucial in controlling legged robots, where stability and optimal control are vital. Many control problems can be formulated as convex optimization problems, with a convex cost function and constraints capturing system dynamics. Our review focuses on active balancing problems and presents a general framework for formulating them as second-order cone programming (SOCP) for robustness and efficiency with existing interior point algorithms. We then discuss some prior work around the Zero Moment Point stability criterion, Linear Quadratic Regulator Control, and then the feedback model predictive control (MPC) approach to improve prediction accuracy and reduce computational costs. Finally, these techniques are applied to stabilize the robot for jumping and landing tasks. Further research in convex optimization of legged robots can have a significant societal impact. It can lead to improved gait planning and active balancing which enhances their ability to navigate complex environments, assist in search and rescue operations and perform tasks in hazardous environments. These advancements have the potential to revolutionize industries and help humans in daily life.

Optimization and Control,Robotics

What problem does this paper attempt to address?

The paper attempts to address the problem of achieving efficient and stable control in legged robots. Specifically, the authors focus on solving the active balancing problem of legged robots through convex optimization methods. These issues include:

1. **Stability Control**: How to ensure the stability and balance of legged robots in various dynamic environments, especially in high-dynamic tasks such as jumping and landing.
2. **Optimizing Control Inputs**: How to generate optimal control inputs to minimize deviations from the preset trajectory while satisfying system dynamics and physical constraints.
3. **Real-time Performance**: How to achieve real-time control and prediction with limited computational resources to cope with uncertainties and disturbances in the environment.

### Main Research Content

1. **Zero Moment Point (ZMP) Criterion**:
- ZMP is a method used to analyze the stability of legged robots by ensuring that the projection point of all forces and moments acting on the robot on the ground is zero, thereby maintaining the robot's balance.
- The ZMP criterion is simple and intuitive, suitable for real-time monitoring and adapting to environmental changes, but its assumptions are relatively simplified and may not be applicable to complex real-world environments.

2. **Linear Quadratic Regulator (LQR) Control**:
- LQR calculates control inputs by optimizing a quadratic cost function to minimize deviations from the desired state, ensuring stability and smooth motion.
- LQR control is simple and easy to implement, but it is mainly suitable for linear systems and has limited capability in handling nonlinear systems, requiring an accurate dynamic model.

3. **Model Predictive Control (MPC)**:
- MPC calculates optimal control inputs by predicting the future behavior of the system and minimizing a cost function within a limited time horizon.
- MPC can handle nonlinear systems and uncertainties in the environment, but it has high computational complexity, especially for large and complex legged robots.

4. **Sequential Linear Quadratic Model Predictive Control (SLQ-MPC)**:
- SLQ-MPC achieves stability control by linearly approximating nonlinear dynamics and calculating linear feedback control laws at each time step.
- This method combines the optimal control of MPC with the computational efficiency of linear approximation, making it suitable for complex legged robots.

5. **Nonlinear Model Predictive Control (NMPC)**:
- NMPC can more effectively handle nonlinear systems and constraints, providing more precise and efficient robot motion, but it has high computational complexity and requires powerful computational resources.

6. **Active Balancing Problem**:
- The authors propose a method to transform the active balancing problem into a Second-Order Cone Programming (SOCP) problem by introducing new variables and constraints, making the problem have a global optimal solution.
- The SOCP method utilizes known efficient interior-point algorithms and can achieve good performance in different types of active balancing problems.

### Research Significance

Through these methods, legged robots can achieve more stable and efficient motion in complex environments, enhancing their application potential in fields such as search and rescue and hazardous environment operations. The advancements in these technologies not only help improve the autonomy and adaptability of robots but also provide important theoretical foundations and technical support for future robot design and control.