Model Predictive Control for Autonomous Driving considering Actuator Dynamics (1803.03478v2)

Abstract: In this paper, we propose a new model predictive control (MPC) formulation for autonomous driving. The novelty of our MPC stems from the following results. Firstly, we adopt an alternating minimization approach wherein linear velocities and angular accelerations are alternately optimized. We show that in contrast to the joint optimization, the alternating minimization exploits the structure of the problem better, which in turn translates to reduction in computation time. Secondly, our MPC explicitly incorporates the time dependent non-linear actuator dynamics that captures the transient response of the vehicle for a given commanded velocity. This added complexity improves the predictive component of MPC resulting in improved margin of inter-vehicle distance during maneuvers like overtaking, lane-change, etc. Although, past works have also incorporated actuator dynamics within MPC, there has been very few attempts towards coupling actuator dynamics to collision avoidance constraints through the non-holonomic motion model of the vehicle and analyzing the resulting behavior. We use a high fidelity simulator to benchmark our actuator dynamics augmented MPC with other related approaches in terms of metrics like inter-vehicle distance, trajectory smoothness, and velocity overshoot.

• The paper presents an MPC framework that integrates alternating minimization to decouple velocity and angular acceleration optimization.
• It incorporates a data-driven first-order actuator model to map commanded to actual velocities, enhancing vehicle response and safety.
• Simulations in CARSIM demonstrate a 17% reduction in computation time and a 32% decrease in iterations compared to traditional MPC methods.

Model Predictive Control for Autonomous Driving considering Actuator Dynamics

This paper explores a novel Model Predictive Control (MPC) formulation designed for autonomous driving, focusing specifically on addressing actuator dynamics and improving optimization frameworks. By incorporating an alternating minimization approach, the authors aim to exploit the structural nature of the problem to enhance computational efficiency and predictive accuracy.

MPC Framework and Optimization Techniques

The researchers introduce an alternating minimization (AM) methodology where the optimization alternates between linear velocities and angular accelerations. This bifurcation into layers allows each optimization step to address the convex structure more effectively, particularly when dealing with non-holonomic constraints. The main advantage of the AM strategy is its ability to search within a space defined by angular accelerations and velocities separately, thereby reducing the computational burden compared to joint optimization strategies.

Additionally, the paper explores the actuator dynamics, emphasizing a time-dependent non-linear mapping between commanded and actual vehicle velocities. By modeling the actuator as a first-order system and fitting parameters using data-driven methods, the authors demonstrate how this explicit incorporation within the MPC leads to improved vehicle dynamics, such as maintaining inter-vehicle distance during complex maneuvers like lane changes and overtaking.

Implementation and Benchmarking

The proposed MPC framework was validated through simulations using a high-fidelity vehicle simulator, CARSIM. These simulations showcased the formulation's efficacy in reducing computational time by 17% and the number of iterations needed by 32% compared to conventional MPC methodologies without the described structural optimizations.

As part of their experimental setup, constraints such as velocity and angular acceleration bounds were adhered to, with the AM model showing tangible improvements in scenarios involving collision avoidance and trajectory smoothness. Comparisons were made between models incorporating the actuator dynamics and those with simpler velocity assumptions, with results indicating better transient response and improved damping characteristics with the former approach.

Practical and Theoretical Implications

From a practical standpoint, the research highlights how considering actuator dynamics in control frameworks could lead to more responsive and anticipatory autonomous vehicles. The substantial reduction in required computation time makes the approach suitable for real-time applications.

Theoretically, this work reinforces the importance of coupling realistic actuator models with optimization frameworks. By leveraging the AM approach, the research provides a promising direction for developing MPC frameworks that are both computationally efficient and robust in handling complex vehicle dynamics.

Future Directions

Future research could expand on this work by exploring more complex actuator models, including possible second-order dynamics or steering actuator profiles. Implementing these frameworks in real-world autonomous platforms will be crucial for validating the methods beyond simulated environments.

In conclusion, the paper offers a detailed exploration into the integration of actuator dynamics within an MPC framework for autonomous driving, demonstrating improved computational efficiency while maintaining essential trajectory quality and safety metrics.