A decision control method for autonomous driving based on multi-task reinforcement learning

In contrast to the traditional pneumatic braking system, the electronic-controlled pneumatic braking system of commercial vehicles is a new system and can remedy the defects of the conventional braking system, such as long response time and low control accuracy. Additionally, it can adapt to the needs and development of autonomous driving. As the key pressure regulating component in electronic-controlled pneumatic braking system of commercial vehicles, automatic pressure regulating valves can quickly and accurately control the braking pressure in real time through an electronic control method. By aiming at improving driving comfort on the premise of ensuring braking security, this paper took the automatic pressure regulating valve as the research object and studied the pressure change rate during the braking process. First, the characteristics of the automatic pressure regulating valve and the concept of the pressure change rate were elaborated. Then, with the volume change of automatic pressure regulating valve in consideration, the mathematical model based on gas dynamics and the association model between pressure change rate and vehicle dynamic model was established in MATLAB/Simulink and analyzed. Next, through the experimental test of a sample product, the mathematical models have been verified. Finally, the key structure parameters affecting the pressure change rate of the automatic pressure regulating valve and the influence law have been identified; therefore, appropriate design advice and theoretical support have been provided to improve driving comfort.

Download Full-text

Independent wheel control system design for highly automated driving

Proceedings of the Institution of Mechanical Engineers Part D Journal of Automobile Engineering ◽

10.1177/09544070211006525 ◽

2021 ◽

pp. 095440702110065

Author(s):

Shihuan Li ◽

Lei Wang

Keyword(s):

Control System ◽

Control Method ◽

Autonomous Driving ◽

Control System Design ◽

Steering Control ◽

Automated Driving ◽

Accuracy And Precision ◽

Actual Observation ◽

Control Scheme ◽

Interference Measurement

For L4 and above autonomous driving levels, the automatic control system has been redundantly designed, and a new steering control method based on brake has been proposed; a new dual-track model has been established through multiple driving tests. The axle part of the model was improved, the accuracy of the transfer function of the model was verified again through acceleration-slide tests; a controller based on interference measurement was designed on the basis of the model, and the relationships between the controller parameters was discussed. Through the linearization of the controller, the robustness of uncertain automobile parameters is discussed; the control scheme is tested and verified through group driving test, and the results prove that the accuracy and precision of the controller meet the requirements, the robustness stability is good. Moreover, the predicted value of the model fits well with the actual observation value, the proposal of this method provides a new idea for avoiding car out of control.

Download Full-text

Transfer Reinforcement Learning for Autonomous Driving

ACM Transactions on Modeling and Computer Simulation ◽

10.1145/3449356 ◽

2021 ◽

Vol 31 (3) ◽

pp. 1-26

Author(s):

Aravind Balakrishnan ◽

Jaeyoung Lee ◽

Ashish Gaurav ◽

Krzysztof Czarnecki ◽

Sean Sedwards

Keyword(s):

Decision Making ◽

Reinforcement Learning ◽

Transfer Problem ◽

Autonomous Driving ◽

High Fidelity ◽

Rule Based ◽

High Level ◽

Real Vehicle

Reinforcement learning (RL) is an attractive way to implement high-level decision-making policies for autonomous driving, but learning directly from a real vehicle or a high-fidelity simulator is variously infeasible. We therefore consider the problem of transfer reinforcement learning and study how a policy learned in a simple environment using WiseMove can be transferred to our high-fidelity simulator, W ise M ove . WiseMove is a framework to study safety and other aspects of RL for autonomous driving. W ise M ove accurately reproduces the dynamics and software stack of our real vehicle. We find that the accurately modelled perception errors in W ise M ove contribute the most to the transfer problem. These errors, when even naively modelled in WiseMove , provide an RL policy that performs better in W ise M ove than a hand-crafted rule-based policy. Applying domain randomization to the environment in WiseMove yields an even better policy. The final RL policy reduces the failures due to perception errors from 10% to 2.75%. We also observe that the RL policy has significantly less reliance on velocity compared to the rule-based policy, having learned that its measurement is unreliable.

Download Full-text