This paper focuses on the real-time trajectory planning problem for autonomous vehicles driving in realistic urban environments. To solve the complex navigation problem, we adopt a hierarchical motion planning framework. First, a rough reference path is extracted from the digital map using commands from the high-level behavioral planner. The conjugate gradient nonlinear optimization algorithm and the cubic B-spline curve are employed to smoothen and interpolate the reference path sequentially. To follow the refined reference path as well as handle both static and moving objects, the trajectory planning task is decoupled into lateral and longitudinal planning problems within the curvilinear coordinate framework. A rich set of kinematically feasible path candidates are generated to deal with the dynamic traffic both deliberatively and reactively. In the meanwhile, the velocity profile generation is performed to improve driving safety and comfort. After that, the generated trajectories are carefully evaluated by an objective function, which combines behavioral decisions by reasoning about the traffic situations. The optimal collision-free, smooth, and dynamically feasible trajectory is selected and transformed into commands executed by the low-level lateral and longitudinal controllers. Field experiments have been carried out with our test autonomous vehicle on the realistic inner-city roads. The experimental results demonstrated capabilities and effectiveness of the proposed trajectory planning framework and algorithms to safely handle a variety of typical driving scenarios, such as static and moving objects avoidance, lane keeping, and vehicle following, while respecting the traffic rules.

Accurate maneuver prediction for surrounding vehicles enables intelligent vehicles to make safe and socially compliant decisions in advance, thus improving the safety and comfort of the driving. The main contribution of this paper is proposing a practical, high-performance, and low-cost maneuver-prediction approach for intelligent vehicles. Our approach is based on a dynamic Bayesian network, which exploits multiple predictive features, namely, historical states of predicting vehicles, road structures, as well as traffic interactions for inferring the probability of each maneuver. The paper also presents algorithms of feature extraction for the network. Our approach is verified on real traffic data in large-scale publicly available datasets. The results show that our approach can recognize the lane-change maneuvers with an F1 score of 80% and an advanced prediction time of 3.75 s, which greatly improves the performance on prediction compared to other baseline approaches.

Path planning is an essential function in an intelligent vehicle, especially when driving in scenarios cluttered by large-scale static obstacles. Traditional path planners often struggle to find a balance among speed, accuracy, and optimality in their solutions. In this paper, we introduce an Adaptive Pure Pursuit (APP) planner, which is designed to be fast and near-optimal for autonomous driving in cluttered environments. The APP planner generates feasible paths through a simulated closed-loop tracking control process of a virtual vehicle. If a derived path encounters obstacles, an adaptive refinement step is taken to locally reduce these collisions. Unlike search-based planners that suffer from the “curse of dimensionality” and optimization-based methods that often run slowly, the APP planner operates extremely fast. The high speed stems from the fact that both the virtual controller simulation and the refinement step involve computations with zero degrees of freedom. The proposed APP planner outperforms the prevalent optimization-based and search-based path planners, as shown by comparative simulations. Real-world experiments were also conducted to validate the APP planner, and its source codes are provided at https://github.com/libai1943/Adaptive_Pure_Pursuit_Planner .

In this paper, we develop an integrated local trajectory planning and control scheme for the navigation of autonomous ground vehicles (AGVs) along a reference path with avoidance of static obstacles. Instead of applying traditional cross track-based feedback controllers to steer the vehicle to track the reference path as closely as possible, we decompose the path following task into two subtasks. Firstly, in order to follow the reference path with smooth motions and avoid obstacles as well, we apply an efficient model-based predictive trajectory planner, which considers geometric information of the desired path, kinematic constraints and partial-dynamic constraints to obtain a collision-free, and dynamically-feasible trajectory in each planning cycle. Then, the generated trajectory is fed to the low-level trajectory tracking controller. Relying on the steady-state steering characteristics of vehicles, we develop an internal model controller to track the desired trajectory, while rejecting the negative effects resulting from model uncertainties and external disturbances. Simulation results demonstrate capabilities of the proposed algorithm to smoothly follow a reference path while avoiding static obstacles at a high speed.

This paper presents a novel cooperative trajectory planning approach for semi-autonomous driving. The machine interacts with the driver at the decision level and the trajectory generation level. To minimize conflicts between the machine and the human, the trajectory planning problem is decomposed into a high-level behavior decision-making problem and a low-level trajectory planning problem. The approach infers the driver’s behavioral semantics according to the driving context and the driver’s input. The trajectories are generated based on the behavioral semantics and driver’s input. The feasibility of the proposed approach is validated by real vehicle experiments. The results prove that the proposed human–machine cooperative trajectory planning approach can successfully help the driver to avoid collisions while respecting the driver’s behavior.

Achieving reliable and safe autonomous driving in off-road environments requires accurate and efficient terrain traversability analysis. However, this task faces several challenges, including the scarcity of large-scale datasets tailored for off-road scenarios, the high cost and potential errors of manual annotation, the stringent real-time requirements of motion planning, and the limited computational power of onboard units. To address these challenges, this paper proposes a novel traversability learning method that leverages self-supervised learning, eliminating the need for manual annotation. We adopt bird's-eye view (BEV) as the input representation to better support ego-centric coordinate representation and integration with motion planning. During vehicle operation, the proposed method conducts online analysis of traversed regions and dynamically updates prototypes to adaptively assess the traversability of the current environment, effectively handling dynamic scene changes. We evaluate our approach against state-of-the-art benchmarks on both public datasets and our own dataset, covering diverse seasons and geographical locations. Experimental results demonstrate that our method achieves superior performance over recent approaches in terms of Area Under the Receiver Operating Characteristic curve (AUROC), Average Precision (AP), and F1-score. Additionally, real-world vehicle experiments show that our method operates at 10 Hz, while a 5.5 km autonomous driving experiment further validates the generated traversability cost map's compatibility with downstream motion planning.