Positioning technology is an important component of environmental perception and the basis of decision-making and motion control for intelligent vehicles. Current independent positioning methods inevitably have defects and poor environmental adaptability. To improve positioning accuracy and stability using low-cost sensors, a multi-sensor fusion positioning strategy for intelligent vehicles using global pose graph optimization is proposed in this paper. This strategy complements the advantages of multiple sensors to obtain global optimal positioning results, and it can cope with a variety of challenging working conditions. Considering the fast response of the inertial measurement unit (IMU) and the advantages of a camera without drift, this strategy realizes the tightly coupled fusion of vision and the IMU to obtain the optimal pose of visual-inertial odometry (VIO) by nonlinear optimization. Moreover, this strategy refines the modeling of the IMU, and proposes a keyframe selection strategy to deal with driving in a straight line without obvious feature changes, turning, and other driving conditions. Taking into account the versatility of the global positioning system (GPS) and the low cost of the network real-time kinematic (NRTK) positioning, the output of VIO and GPS/NRTK is adaptively fused through global pose graph optimization to maintain global consistency. Finally, the multi-sensor fusion positioning strategy is verified using the KITTI dataset on the ROS simulation platform. The results show that the multi-sensor fusion positioning strategy for intelligent vehicles has better accuracy and environmental adaptability at a lower cost than state-of-the-art strategies.

Intelligent vehicle positioning is an enabling technology for decision-making, trajectory planning, and motion control, and is a key technology in intelligent transportation systems (ITSs). In order to obtain a low-cost and high-precision positioning strategy, an integrated positioning strategy for intelligent vehicles based on the global navigation satellite system (GNSS), dead reckoning (DR), ultra-wide band (UWB), and visual map matching (VMM) was proposed in this paper. First, the error sources of the three independent positioning methods of GNSS, DR, and UWB were analyzed, and independent positioning algorithms with improved accuracy were proposed. Then, the fusion positioning algorithm of GNSS/DR/UWB with an adaptive information distribution coefficient was established using a federated Kalman filter to improve the positioning accuracy and continuity. Afterward, to avoid reliance on the previous moment, the positioning result was further corrected by VMM. Finally, offline simulation and real vehicle tests were conducted under typical working conditions with the impact of real-world noise and the real-motion states of vehicles. The results showed that the GNSS/UWB/DR/VMM positioning algorithm could effectively improve the positioning accuracy and reliability of intelligent vehicles at a low cost.

This study addresses the virtual testing of intelligent driving, examines the key problems in modeling and simulating millimeter wave radar environmental clutter, and proposes a modeling and simulation method for the environmental clutter of millimeter wave radar in intelligent driving. First, based on the attributes of intelligent vehicle millimeter wave radar, the classification characteristics of the traffic environment of an intelligent vehicle and the generation mechanism of radar environmental clutter are analyzed. Next, the statistical distribution characteristics of the clutter amplitude, the distribution characteristics of the power spectrum, and the electromagnetic dielectric characteristics are analyzed. The simulation method of radar clutter under environmental conditions such as road surface, rainfall, snowfall, and fog are deduced and designed. Finally, experimental comparison results are utilized to validate the model and simulation method.

Dynamic scheduling of access to shared resources by autonomous systems is a challenging problem, characterized as being NP-hard. The complexity of this task leads to a combinatorial explosion of possibilities in highly dynamic systems where arriving requests must be continuously scheduled subject to strong safety and time constraints. An example of such a system is an unsignalized intersection, where automated vehicles' access to potential conflict zones must be dynamically scheduled. In this paper, we apply Neural Monte Carlo Tree Search (NMCTS) to the challenging task of scheduling platoons of vehicles crossing unsignalized intersections. Crucially, we introduce a transformation model that maps successive sequences of potentially conflicting road-space reservation requests from platoons of vehicles into a series of board-game-like problems and use NMCTS to search for solutions representing optimal road-space allocation schedules in the context of past allocations. To optimize search, we incorporate a prioritized re-sampling method with parallel NMCTS (PNMCTS) to improve the quality of training data. To optimize training, a curriculum learning strategy is used to train the agent to schedule progressively more complex boards culminating in overlapping boards that represent busy intersections. In a busy single four-way unsignalized intersection simulation, PNMCTS solved 95\% of unseen scenarios, reducing crossing time by 43\% in light and 52\% in heavy traffic versus first-in, first-out control. In a 3x3 multi-intersection network, the proposed method maintained free-flow in light traffic when all intersections are under control of PNMCTS and outperformed state-of-the-art RL-based traffic-light controllers in average travel time by 74.5\% and total throughput by 16\% in heavy traffic.