TL;DR
This paper introduces a new Robust Model Predictive Control (RMPC) strategy that combines a stable nominal controller with an optimization-based auxiliary controller using Crisscross Optimization (CSO). This approach significantly improves trajectory tracking performance by reducing conservatism and enhancing transient response compared to traditional RMPC methods.
What Problem Does This Paper Address?
Trajectory tracking is crucial for many modern systems, from autonomous vehicles to industrial robots. However, real-world systems are subject to uncertainties and disturbances, making precise tracking challenging. Traditional Robust Model Predictive Control (RMPC) methods are designed to handle these uncertainties and guarantee stability, but they often suffer from excessive conservatism. This conservatism leads to slower response times and suboptimal performance, hindering the system's ability to achieve the desired trajectory quickly and accurately. The paper aims to overcome this trade-off between robustness and performance by developing an RMPC strategy that maintains stability while actively improving dynamic performance.
Methods
The proposed method introduces a composite control structure. First, a nominal controller ($u_c(k)$) is designed using a min-max optimization approach, which guarantees the closed-loop stability of the system. This nominal controller is derived from Linear Matrix Inequality (LMI) conditions to ensure robustness against uncertainties.
The key innovation lies in the second stage: an auxiliary controller ($u_a(k)$) is introduced to enhance transient performance. This auxiliary controller is optimized using the Crisscross Optimization (CSO) algorithm. CSO is an intelligent optimization method inspired by biological organisms, known for its rapid convergence and ease of implementation. It performs global optimization by iteratively refining a population of potential solutions through horizontal and vertical crossover operations. The CSO algorithm searches for the optimal auxiliary control input within a feasible range, effectively "fine-tuning" the control action to achieve faster and more accurate tracking without compromising the stability guaranteed by the nominal controller. The overall control input is the sum of the nominal and auxiliary control inputs: $u(k) = u_c(k) + u_a(k)$.
Datasets
Not discussed in the paper. The paper uses simulated data from an angular positioning system.
Key Insights and Ideas
- De-conservatizing RMPC: The core idea is to separate the robustness guarantee (handled by the nominal controller) from the performance enhancement (handled by the auxiliary controller), allowing for a more aggressive and responsive control action.
- CSO for Performance Tuning: Crisscross Optimization is shown to be an effective tool for fine-tuning control parameters within a stable region, leading to significant improvements in transient performance metrics like rise time and settling time.
- Composite Control Structure: The combination of a theoretically guaranteed stable nominal controller with an empirically optimized auxiliary controller offers a powerful approach to balance robustness and performance.
- Flexibility in Performance Objectives: The CSO's objective function can be formulated more flexibly than traditional RMPC cost functions, allowing for the inclusion of terms that specifically target tracking speed and error reduction.
Key Results
The proposed CSO-based RMPC strategy was validated on an angular positioning system for two trajectory tracking scenarios: constant set-point tracking and ramp trajectory tracking.
- Constant Set-Point Tracking:
- Reduced rise time by approximately 23.5% (1.3s vs. 1.7s).
- Reduced settling time by approximately 27.8% (1.3s vs. 1.8s).
- Demonstrated a more aggressive control action that accelerates tracking response while ensuring smooth convergence.
- Ramp Trajectory Tracking:
- Achieved tracking within approximately 0.8s, compared to nearly 1.2s for conventional RMPC.
- Showed a faster and smoother tracking response.
Overall, the results indicate that the proposed method significantly improves tracking dynamics and control performance compared to conventional RMPC, effectively mitigating conservatism.
Limitations and Future Work
The authors acknowledge two main limitations:
- Empirical Tuning of $K_2$ Range: The admissible range for the auxiliary control gain ($K_2$) is currently determined empirically, lacking systematic theoretical guidance.
- Computational Burden of Online Optimization: The real-time online optimization performed by CSO can impose a significant computational burden, making it less suitable for high-dimensional systems.
Future work will focus on:
- Developing systematic theoretical guidance for determining the $K_2$ range.
- Addressing the computational challenges of online optimization for high-dimensional systems.
- Extending the proposed method to handle more complex systems with parametric uncertainties, time delays, and bounded disturbances.
Summary updated