TL;DR
This paper introduces a novel robust tracking controller for nonlinear systems that guarantees exponential convergence of the system's state to a desired trajectory, even in the presence of unknown, matched uncertainties. By combining control contraction metrics (CCM) with a disturbance estimation technique, the controller achieves superior performance compared to existing methods, as demonstrated on aircraft and quadrotor systems.
What Problem Does This Paper Address?
Many real-world robotic systems, such as aircraft and drones, are inherently nonlinear and subject to uncertainties (e.g., unmodeled dynamics, external forces) and disturbances. Designing controllers that can accurately track desired trajectories in the face of these uncertainties is a significant challenge. Traditional robust control methods can be overly conservative, while adaptive methods often require specific parametric models for the uncertainties. Existing control contraction metric (CCM) approaches, while powerful for nominal systems, need to be extended to handle these uncertainties effectively while still guaranteeing strong convergence properties like exponential stability. This paper aims to bridge this gap by developing a controller that robustly handles matched uncertainties and ensures exponential convergence.
Methods
The core of the proposed approach lies in the synergistic combination of two key techniques:
- Control Contraction Metrics (CCM): CCM provides a framework for designing controllers that ensure exponential convergence of system states. It extends contraction theory, which analyzes how distances between trajectories evolve, to the control design problem. The paper leverages existing CCM theory to find a suitable "metric" for the system dynamics.
- Disturbance Estimation: To handle the unknown uncertainties, the authors employ a disturbance estimation scheme. This scheme estimates the pointwise value of the uncertainty at each time step. Crucially, it also provides a pre-computable estimation error bound (EEB), which quantifies the maximum possible error in the disturbance estimate.
These two components are integrated into a "Robust Riemannian Energy Condition" (RRE). This condition, derived from the CCM framework, is modified to account for the estimated disturbance and its associated error bound. The controller is then synthesized by solving a quadratic programming (QP) problem that minimizes control effort while satisfying this RRE condition. This ensures that the system's "Riemannian energy" (a measure of distance from the desired trajectory) decreases exponentially, even with the estimated disturbance.
Datasets
Not discussed in the paper. The paper uses simulated examples of an aircraft and a planar quadrotor for evaluation.
Key Insights and Ideas
- Bridging Robustness and Exponential Convergence: The paper demonstrates that by explicitly estimating the disturbance and its error bound, one can modify the CCM framework to achieve robust exponential convergence, a stronger guarantee than asymptotic convergence often provided by other robust or adaptive methods.
- Disturbance Estimation with Error Bounds is Key: The ability to provide a quantifiable bound on the disturbance estimation error (EEB) is critical. This EEB allows the controller to conservatively account for the uncertainty, ensuring stability even when the exact disturbance is unknown.
- CCM as a Foundation for Robustness: The paper shows that a CCM designed for the nominal system (without uncertainties) can still be used as a basis for designing a robust controller for the uncertain system, provided the disturbance estimation and RRE condition are incorporated.
- Practical Implications for Robotics: The successful application to aircraft and quadrotor models suggests this approach could be valuable for real-world robotic systems where precise tracking is essential and uncertainties are unavoidable.
Key Results
- Aircraft Example: The proposed DE-CCM controller outperformed a standard CCM (which ignores uncertainties) and an adaptive CCM (Ad-CCM) in terms of tracking performance. It achieved a 54% lower Mean Squared Error (MSE) than CCM and a 2% lower MSE than Ad-CCM. The control inputs were also observed to be smoother than those from Ad-CCM.
- Planar Quadrotor Example: DE-CCM again showed superior performance, achieving a 46% lower MSE than CCM and a 14% lower MSE than Ad-CCM. The controller also successfully navigated the quadrotor around obstacles, whereas the CCM controller collided with one.
- Theoretical Guarantees: The paper provides a theoretical guarantee of universal exponential stabilization for the uncertain system under certain assumptions.
Limitations and Future Work
- Matched Uncertainties: The current controller is designed for matched uncertainties, meaning the uncertainties enter the system dynamics through the same channels as the control inputs. Many practical systems exhibit unmatched uncertainties, which are not addressed by this work.
- Verification of Control Input Bounds: While the controller guarantees exponential convergence, verifying that the generated control inputs remain within the system's actuator limits (U) is noted as a challenge and an area for future work.
- Continuous-Time Assumption: The theoretical guarantees are based on a continuous-time implementation. In practice, controllers run on digital processors with finite sampling times, which can slightly violate the exponential convergence property.
- Computational Cost: The geodesic computation step for the CCM can be computationally intensive, although the authors suggest it's more manageable than nonlinear model predictive control.
- Experimental Validation: The paper focuses on simulations. The authors suggest future work should include experimental validation of the proposed controller on real hardware.
- Addressing Unmatched Uncertainties: A key future direction is to extend the methodology to handle unmatched uncertainties.
Summary updated