Section: New Results

Plug&Play control for highly non-linear systems: Stability analysis of autonomous vehicles

Participants : Francisco Navas, Fawzi Nashashibi.

The final stage for automating a vehicle relies on the control algorithms. They are in charge of providing the proper behavior and performance to the vehicle, leading to provide fully automated capabilities. Controllability and stability of dynamic complex systems are the key aspects when it comes to design intelligent control algorithms for vehicles.

Nowadays, the problem is that control systems are “monolithic”. That means that a minor change in the system could require the entire redesign of the control system. It addresses a major challenge, a system able to adapt the control structure automatically when a change occurred.

An autonomous vehicle is built by combining a set-of-sensors and actuators together with sophisticated algorithms. Since sensors and actuators are prone to intermittent faults, the use of different sensors is better and more cost effective than duplicating the same sensor type. The problem is to deal with the different availability of each sensor/actuator and how the vehicle should react to these changes. Another possible modification is the change in vehicle dynamics over time; or difference in dynamics from one vehicle to another.

A methodology that improves the security of autonomous driving systems by providing a framework managing different dynamics and sensor/actuator setups should be carried out. New trends are proposing intelligent algorithms able to handle any unexpected circumstances as unpredicted uncertainties or even fully outages from sensors. This is the case of Plug & Play control, which is able to provide stability responses for autonomous vehicles under uncontrolled circumstances.

Here, the basis of Plug & Play control, Youla-Kucera parameterization, has been used to develop different applications within the autonomous driving field.

• Stable controller reconfiguration when some change occurs. Last year, the already commercially available Adaptive Cruise Controller (ACC) system, and its evolution by adding vehicle-to-vehicle communication (CACC) were examined. The Youla-Kucera parameterization was used for providing stable transitions between both controllers when the vehicle-to-vehicle communication link is changing from available to disable or vice-versa. More details can be found in [52]. This year, this work has been extended in what is called Youla-Kucera-based Advanced Cooperative Adaptive Cruise Control (ACACC). In the literature, CACC degrades to ACC when communication when the preceding vehicle is no longer available. This degradation occurs even if information from another V2V-equipped vehicle ahead (different from the preceding vehicle) is still available. ACACC benefits from the existing communication with this vehicle ahead in the string, reducing the inter-vehicle distance whereas keeping string stability. The proposed structure uses YK parameterization to obtain a hybrid behavior between two CACC controllers with different time gaps. Stable transition between both controllers is also ensured. This work has been submitted to IEEE Transactions on Vehicular Technology. Finally, Youla-Kucera has been also employed to assure stable transitions when other CACC-equipped vehicles are joining/leaving a CACC string of vehicles.

• Online closed loop identification. Youla-Kucera has a dual formulation that allows recasting closed-loop identification into open-loop-like identification. [28] deals with the identification of longitudinal dynamics of a cycab for subsequent control performance's improvement. Here, the dual Youla-Kucera formulation is used to transform a closed-loop identification problem in an open-loop-like. The algorithm is tested in a string of two cycabs equipped with a proportional-derivative-based CACC, showing how the resulting model is improved in comparison with a classical open-loop identification algorithm. Closed-loop identification results have been also obtained for a production vehicle when connected to a lane following control system. Thanks to that, lateral dynamics are known for velocities between 8 and $20m/s$.

• A final step that integrates both stable controller reconfiguration and closed-loop identification: Automatic control reconfiguration to achieve optimal performance based on the identification of the new situation. This idea has been used to obtain an adaptive approach able to ensure string stability when different dynamics are involved in the same string of vehicles (a heterogeneous string of vehicles). A supervisor is able to provide the closest model in a predefined set, activating the controller that ensures string stability. The closest model in the set can be known without using identification algorithms, thanks to Youla-Kucera properties, with the consequent computational saving.