Ensuring adaptive safe control performance in the persistence of uncertainties poses great challenges for safety-critical autonomous vehicle control systems. This paper investigates the adaptive ...safety control design for the trajectory tracking problem on a four-wheel-drive electric vehicle (4WDEV) by employing adaptive control barrier functions (CBFs) to formulate the state constraints of the affine control system. Specifically, the tracking-error model is first established with consideration of integrated longitudinal and lateral dynamics of the 4WDEV. Then, adaptive CBFs and control Lyapunov functions are employed to formulate the safety constraints and stability conditions of the constrained control problem, respectively. Through the safety control design, the trajectory tracking control problem is constructed as quadratic programming. The adaptive performance is enhanced by introducing the time-varying parameters for CBFs to be adjusted by an auxiliary system, which also enhances the feasibility of the quadratic optimization. The effectiveness of the proposed safety control design is verified though simulations of different scenarios and hardware-in-the-loop experiments. Through comparisons among the conventional high-order CBFs and baseline controllers, e.g., linear quadratic regulator and model predictive control, the superiority of the proposed control design in the aspect of safety satisfaction and adaptability are effectively demonstrated.
Model Predictive Path Integral (MPPI) is a recognized sampling-based approach for finite horizon optimal control problems. However, the efficacy and computational efficiency of prevailing MPPI ...methods are heavily reliant on the quality of rollouts. This is problematic because it is hard to sample a low-cost trajectory using random control sequences, thereby leading to inferior performance and computational efficiency, especially under constrained resources. To address this issue, we propose a data-efficient MPPI method called reinforcement learning-driven MPPI (RL-driven MPPI), which significantly reduces the dependency on the quantity and quality of samples. RL-driven MPPI employs an offline-online policy learning scheme, where the offline policy learned by RL serves as the initial solution and the initial rollout generator of MPPI, effectively combining the strengths of both RL and MPPI. The rollouts generated by RL typically correspond to a lower cost-to-go compared to random sampling, which significantly boosts the sample efficiency and convergence speed of MPPI. Moreover, the value function learned by RL offers an accurate estimation for infinite-horizon cost-to-go, enabling it to serve as a terminal term for the cost criteria of MPPI. This approach empowers MPPI to approximate an infinite-horizon cost with a shorter prediction horizon, thus enhancing real-time performance at each time step. An unmanned aerial vehicle control task is conducted to evaluate the proposed method. Results indicate that the proposed RL-driven MPPI method exhibits superior control performance and sample efficiency.
The friction coefficient identification of dry clutch is a precondition for precise model-based dry clutch control. Recursive Least Square with forgetting factors is adopted to identify the friction ...coefficient of dry clutch during vehicle starting-up process. Because the accuracy of the engine damp coefficient affects friction coefficient identification of the clutch, it is necessary to identify the clutch friction coefficient and engine damp coefficient simultaneously. Through model simulation, it is shown that accurate and fast identification can be achieved no matter what initial values were chosen. It is also verified that the identification performance is good enough for different launch maneuvers with different throttle operations.
Articulated vehicles are susceptible to instability issues due to their distinct dynamic properties. Most existing control strategies focus on constructing an integrated model, yet an accurate ...parametric model for a complex nonlinear system might be unavailable. To address this, a bilevel control structure is established, with the upper level generating corrective yaw moments and the lower level focusing on control allocation, then data-driven predictive control method is introduced, which relies only on input/output measurements to construct a non-parametric representation of the system, this method is implemented in a receding-horizon manner similar to MPC, incorporating constraints to achieve safe maneuvering. The effectiveness of the proposed controller is presented by simulation results, which further confirm its potential in vehicle dynamics control.
As a critical technology for achieving autonomous driving, reinforcement learning has demonstrated its effectiveness in enabling vehicles to learn decision-making and adapt their actions to the ...surrounding environment. Nevertheless, the safety vulnerabilities inherent in reinforcement learning present substantial hazards to highway safety. An AG2RL(Adept Guide and Guard Reinforcement Learning) is proposed to address this issue in this study. The AG2RL algorithm introduces a guardian to safeguard the exploration of the learning agent. The guardian prevents the agent from running out of lane line boundaries, which is good to achieve faster convergence. The simulation experiments are conducted to evaluate the proposed algorithm, and results show that the proposed algorithm has a low accident rate and a faster convergence rate compared with Soft Actor-Critic algorithm.
The integration of electrification and intelligence is of great significance to alleviating range anxiety of electric vehicles. Predictive cruise control (PCC), which optimizes the longitudinal ...driving strategies by using the upcoming road traffic information, can further improve the vehicle economy. The paper gives the comparison of different solution methods regarding PCC problem of electric vehicles. The car-following optimization problem is formulated as a constrained nonlinear optimization problem. For ease of presentation, the car-following optimization problem is reformulated as a standard form of the optimization problem in continuous time domain. Then, the standard form of the optimization problem is transformed from a continuous form to a discrete form by using Euler method and Gauss pseudospectral method. Two common solution methods, that is dynamic programming and sequential quadratic programming, are used to solve the optimization problem in discrete form. Simulations are performed to demonstrate the comparison of different solution schemes.
Full text
Available for:
GEOZS, IJS, IMTLJ, KILJ, KISLJ, NLZOH, NUK, OILJ, PNG, SAZU, SBCE, SBJE, UILJ, UL, UM, UPCLJ, UPUK, ZAGLJ, ZRSKP
Model Predictive Control (MPC) is a good way to solve Predictive Cruise Control (PCC) problems because of its advantages of rolling optimization framework. The time-domain and space-domain models are ...widely used because of their respective advantages. This paper compares the time-domain and the space-domain vehicle models used for predictive energy-saving control in the MPC framework. And the linear model used in this paper can simplify the solution and improve the calculation speed. The similarities and differences of the driving economy in the two domains are analyzed in detail from theoretical modeling to simulation experiments. The multi-objective optimization control problem was established and the parameter sensitivity analysis was completed. It is obvious to know that the sensitive range can be found by reasonable parameter adjustment. And experimental results show that the vehicle model in the space domain presents excellent speed tracking performance, and the computational efficiency of the model is significantly improved. The time-domain vehicle model has better fuel economy.
This work establishes signal temporal logic specifications for driving tasks that autonomous vehicle should execute when turning at intersections. The specifications are based on traffic regulations ...of China. By encoding signal temporal logic specifications as mixed-integer constraints, optimization problems could be established for path planning. By solving the problems with Gurobi optimizer, safe paths could be obtained. Through this work, it is demonstrated that signal temporal logic is a profound tool for formalizing driving tasks and solving path planning problems in autonomous driving. The feasibility of the proposed method is demonstrated via a numerical experiment.