Behavioral decision-making is an important part of the high-level intelligent driving system of intelligent vehicles, and efficient and safe behavioral decision-making plays an important role in the deployment of intelligent transportation system, which is a hot topic of current research. This paper proposes a deep reinforcement learning (DRL) method based on mixed-state space and driving risk for autonomous vehicle behavior decision-making, which enables autonomous vehicles to make behavioral decisions with minimal instantaneous risk through deep reinforcement learning training. Firstly, based on the various behaviors that may be taken by autonomous vehicles during high-speed driving, a calculation method for autonomous vehicle driving risk is proposed. Then, deep reinforcement learning methods are used to improve the safety and efficiency of behavioral decision-making from the interaction between the vehicle and the driving environment. Finally, the effectiveness of the proposed method is proved by training verification in different simulation scenarios, and the results show that the proposed method can enable autonomous vehicles to make safe and efficient behavior decisions in complex driving environments. Compared with advanced algorithms, the method proposed in this paper improves the driving distance of autonomous vehicle by 3.3%, the safety by 2.1%, and the calculation time by 43% in the experiment.

Self-power sensors provide an effective solution of long-term monitoring for many applications. An obstacle in the path is the common mismatching of the harvested power and the required power since the harvester and the sensor are independently designed without cooperative consideration. This article proposes an energy-autonomous integrated system (EAIS) designed to wirelessly acquire and transmit vibration signals without the need for an external power source. The system employs a piezoelectric cantilever that both harvests energy and senses acceleration, integrating two circuit subsystems: the energy harvesting subsystem and the acceleration sensing subsystem. A self-adaptive time-division multiplexing strategy is adopted for the two subsystems so that they can work alternatively in an energy-self-sufficient way. By utilization of a smart and dedicated circuit design, the acceleration sensing subsystem is capable of acquiring full-wave acceleration signals with complicated frequency components, which is considered as a great advancement in self-powered sensing studies. In particular, the entire system is fully energy-autonomous with the capability of self-starting, self-powering and self-sensing.

Harvesting energy from the door closing process provides a promising solution for powering wireless sensors in intelligent transport systems, however issues, such as low conversion efficiency and operational resistance hinder its practical application. In this study, an energy harvesting system (EHS) is proposed and its basic and optimized designs are compared and analyzed. The system integrates a crank‐slider mechanism, a rack‐and‐pinion transmission and a flywheel energy storage module: the crank‐slider converts the rotational motion of the door into linear motion, the rack‐and‐pinion gears are engaged only at preset positions to minimize operating resistance, and the flywheel captures the kinetic energy quickly during the door closing process. The key parameters (crank length: 19.8–25.8 cm; connecting rod length: 25–31 cm; engagement angle: 26°–44°; flywheel inertia: 0–102.6 kg‐mm ² ) were optimized through orthogonal experiments. The results show that the flexible design improves the energy output by 13% compared with the rigid design (35.96–73.79 mJ/time for 1–6 J input), and the orthogonal analysis shows that the eigenvalues of the parameters are distributed in a balanced way, which verifies the feasibility of eliminating the impact kinetic energy loss through elastic contact. In addition, the flexible design reduces the door closing noise by 3–12 dB. Experimental validation shows that the system is capable of providing minute‐level continuous power supply for low‐power sensors.