TL;DR
This paper tackles the challenge of robustly controlling autonomous vehicles on slippery surfaces, where sliding can lead to loss of control. The authors propose a two-stage approach: first, an adaptive observer to estimate tire cornering stiffness and reconstruct sideslip angles even with disturbances, and second, a robust sliding mode controller to ensure accurate path following despite estimation errors and lateral disturbances.
What Problem Does This Paper Address?
Autonomous vehicles, especially those operating in challenging environments like off-road or in adverse weather, face a significant control problem due to tire sliding. When tires lose grip, the vehicle's actual motion deviates from its intended path, potentially leading to instability and accidents. This phenomenon is exacerbated by lateral disturbances (like bumps or uneven terrain) and the difficulty in accurately measuring critical parameters like sideslip angles and tire cornering stiffness in real-time. Existing control methods often struggle to maintain stability and precise trajectory tracking under these conditions, either by assuming ideal conditions or by relying on high-gain controllers that can lead to jerky movements. This paper aims to develop a control strategy that can effectively mitigate sliding effects and maintain accurate path following even in the presence of unmeasurable disturbances and estimation uncertainties.
Methods
The paper proposes a two-pronged approach:
- Robust Adaptive Observer for Cornering Stiffness Identification: The core of this method is a Luenberger-type observer designed to estimate the tire cornering stiffness ($k_f$, $k_r$). This observer is "robust" and "adaptive," meaning it can handle time-varying lateral disturbances and parameter uncertainties. It uses measurements of side-acceleration and yaw rate to identify these crucial tire properties. The estimated cornering stiffness is then used to reconstruct the vehicle's sideslip angle ($\beta$), which is a key indicator of sliding.
- Robust Sliding Mode Controller based on Chained System Theory: Once the sideslip angle is estimated (albeit with potential inaccuracies due to disturbances and estimation errors), a sliding mode controller is designed. This controller leverages the "chained system" theory, a mathematical framework that allows for the design of nonlinear controllers for complex systems by converting them into a simpler, linear-like structure. The controller is designed to be robust to both the inherent inaccuracies in the reconstructed sliding angles and the external lateral disturbances, ensuring that the vehicle states (lateral deviation and orientation error) remain bounded.
Datasets
Not discussed in the paper. The evaluation is primarily based on simulations and experimental results from a real vehicle.
Key Insights and Ideas
- Cornering Stiffness as a Key to Sliding Estimation: The paper highlights that accurately estimating tire cornering stiffness is fundamental to reconstructing sideslip angles and, consequently, compensating for sliding effects.
- Robustness to Disturbances is Crucial: Simply compensating for sliding isn't enough; the control system must also be resilient to external lateral disturbances that can further destabilize the vehicle.
- Chained System Theory for Nonlinear Control: The application of chained system theory provides a structured way to design controllers for complex vehicle dynamics, allowing for the use of well-established linear control techniques within a nonlinear framework.
- Two-Stage Approach for Practicality: The separation into an identification stage (cornering stiffness and sideslip angle estimation) followed by a robust control stage makes the approach more practical, as it doesn't require perfect real-time knowledge of all parameters.
- Trade-offs in Control: The experimental results suggest that while robust controllers can achieve high accuracy, they might sometimes lead to more "drastic" vehicle motion compared to less robust but smoother controllers.
Key Results
- Cornering Stiffness Identification: The robust adaptive observer successfully identified cornering stiffness values, converging to desired values even in the presence of time-varying lateral disturbances. The identification was shown to be relatively insensitive to initial parameter values.
- Sideslip Angle Reconstruction: The reconstructed sideslip angle showed reasonable accuracy, though some reconstruction errors were observed due to the impact of lateral disturbances.
- Lateral Deviation Control: Simulations and experiments demonstrated that the proposed robust anti-sliding controller significantly reduced lateral deviation compared to a controller without sliding compensation or a controller from prior work (Lenain et al., 2006a). The robust controller achieved lateral accuracy within approximately 10 cm.
- Orientation Error Reduction: The controller also improved orientation control accuracy, though complete elimination of orientation errors was not achieved, which aligns with theoretical expectations.
Limitations and Future Work
- Assumptions on Cornering Stiffness and Velocity: The paper acknowledges that the current adaptive laws assume constant velocity and cornering stiffness. Future work aims to relax these assumptions.
- Model Simplifications: The control strategy is primarily based on a kinematic model, with some consideration for lateral dynamics. A fully dynamic model might be necessary for higher speeds or more extreme conditions.
- Experimental Platform: While tested on a real vehicle, the experimental setup had limitations on allowable velocity (3-15 km/h) due to control loop delays.
- Future Directions: The authors suggest designing adaptive robust observers and controllers that do not rely on the assumption of constant velocity and cornering stiffness, and exploring more general ways to handle these variations. They also mention the potential for using sliding mode fuzzy observers for more robust solutions against sustained disturbances.
Summary updated