Yingjie Deng

The Rapidly-exploring random tree (RRT) is a widely used sampling-based algorithm for robot path planning due to its simple structure and strong exploration capability. However, in practical engineering applications with limited planning time, conventional RRT-based methods may suffer from inefficient target capture, redundant exploration, and unstable path quality. To address these issues, this paper proposes a phototaxis-guided RRT-based path planning algorithm named Phototactic-RRT. Specifically, a ‘phototactic growth’ mechanism is introduced to heuristically guide the tree expansion towards the goal, accelerating the discovery of initial trajectories. Sparse sampling controls tree-node density, while SetForbidPoint suppresses repeated expansion from branches associated with previously discovered feasible paths. Furthermore, an ExtendRewire mechanism utilizing ancestry backtracking is implemented to adjust the local tree structure efficiently. Breaking away from the traditional single-path patching paradigm, the proposed algorithm identifies multiple feasible candidate paths and employs a two-stage backend optimization module to geometrically straighten and smooth the trajectories. In four two-dimensional test scenarios, Phototactic-RRT achieved a 100% planning success rate and the lowest median runtime among the six evaluated planners. Its median path cost was lower than those of RRT ^* , Informed-RRT ^* , Q-RRT ^* , and RRT ^* -Smart in every scenario and remained comparable to BIT ^* , with the relative result depending on the map. A paired 30-run evaluation in a three-dimensional voxel environment further demonstrates the applicability of the proposed planning mechanism beyond planar environments under a fixed sampling budget.

Autonomous underwater vehicles (AUVs) require three‐dimensional (3D) path planning in environments where computational efficiency is as critical as solution quality. Sampling‐based planners, such as the RRT* family, incur a time complexity of that grows with the number of sampling points, creating a bottleneck for real‐time AUV operation. This paper introduces the G‐Sunlight algorithm, a deterministic sampling‐based planner that achieves space and time complexity through two coordinated mechanisms: (1) A conical sampling mechanism is designed for 3D space, which employs the cascaded rotated coordinate system to quickly locate tangent points on the obstacle surface. (2) A grid‐constrained filter limits the number of candidate points retained in each spatial cell, suppressing the redundant sampling that degrades the efficiency of both stochastic and deterministic alternatives. Numerical experiments in 2D and 3D environments validate the superiority of the G‐Sunlight algorithm. Physical experiments on a remotely operated vehicle in a 5 m 8 m pool verify the feasibility of the algorithm in real‐world environments. The open‐source codes are available at https://github.com/breaker123344/LIU3Dsunlight .

Autonomous navigation of unmanned sailboats in real-world oceans faces significant challenges due to the strong coupling among stochastic winds, irregular waves, and time-varying currents. Existing approaches often rely on simplified kinematic models or idealized environmental assumptions and therefore fail to capture the substantial drift forces and hydrodynamic disturbances encountered in rough sea states. To address this issue, we propose a robust deep reinforcement learning framework, termed difficulty-adaptive phasic policy gradient (DA-PPG), integrated with a high-fidelity fluid-structure interaction model that explicitly accounts for relative flow velocities and second-order wave-drift forces. To mitigate sparse rewards and training instability in such chaotic environments, DA-PPG introduces a local A*-guided reward-shaping mechanism and a difficulty-adaptive curriculum learning strategy. Comprehensive simulations in high-fidelity scenarios show that the proposed method outperforms traditional geometric planners and baseline DRL algorithms. In particular, the learned policy maintains smoother and more goal-directed navigation under coupled wind-wave-current disturbances and demonstrates zero-shot generalization by proactively avoiding unseen moving vessels, thereby achieving superior trajectory smoothness, maneuvering efficiency, and navigational safety in complex marine environments.

Sunlight algorithm (Deng and Xu 2025 Trans. Intell. Transp. Syst. 26 20657–68) is a novel sampling-based path planning method that we originally introduced for planar mobile robots, such as unmanned surface vehicles and autonomous ground vehicles. It performs the efficient sampling by simulating the sunlight radiation, ensuring a time complexity of O ( n ) compared with O ( n log n ) for traditional RRT*, which makes it competitive in complex mazes. However, the sunlight algorithm has two crucial drawbacks: (1) sampling at equal intervals of light ray angles can hardly search the optimal path in the scenario with dense small obstacles; (2) the time complexity of O ( n ) holds based on the assumption of limited size of Openset , which lacks rigorous evidence in some cases. To overcome these drawbacks, this paper presents two improvements on the sunlight algorithm. First, an adaptive sampling compensation mechanism is established to ensure the uniformity of sampling and prevent the omission of key tangent points lying along the edges of small obstacles. Second, a strict filtering mechanism is established for adding the sampling candidates to Openset , which ensures only the optimal waypoint exists in its visible zone. By adopting the two mechanisms, the sunlight algorithm is granted the excellent searching capability in the environment with dense small obstacles without degrading the computational complexity. The proposed scheme is tested in both simulation and robot field experiment, which is deployed on an Ackermann mobile robot. The results demonstrate its exceptional effectiveness in navigating cluttered spaces, along with the fastest search speed and shortest path length compared to existing techniques.

This article addresses issues such as vehicle body instability, vibration, and model uncertainties in active suspension systems (ASSs) when vehicles traverse rough terrain, proposing an event-triggered adaptive neural asymptotic control (ETANAC) method with prescribed performance for the active suspension unit (ASU). The proposed method consists of prescribed performance control (PPC), adaptive neural backstepping asymptotic control, event-triggered control (ETC), Nussbaum technique, and the ASU model. The PPC method can rapidly adjust the vehicle body displacement to a stable state. The adaptive neural asymptotic control method with prescribed performance effectively reduces the amplitude of body vibration displacement and acceleration, ensuring that the tracking error asymptotically converges to zero. This method addresses the challenge of coordinating vehicle body displacement control and ride comfort. ETC can reduce the actuator triggering frequency. The Nussbaum technique addresses the issue of unknown control direction in the ASU. This article uses Barbalat’s lemma and Lyapunov methods to analyze the stability of the proposed method, proving the asymptotic convergence of all tracking errors. Finally, numerical simulation experiments are conducted to verify the effectiveness and superiority of the proposed method.

Accurate trajectory forecasting plays a pivotal role in various maritime applications, including route optimization, collision prevention, and intelligent traffic management. Traditional approaches, including statistical models and conventional machine learning methods, have demonstrated constrained capabilities in modeling the complex spatiotemporal characteristics of maritime trajectories. Deep learning architectures have shown remarkable potential in processing voluminous navigation data and learning sophisticated movement patterns through their hierarchical feature extraction mechanisms. This study presents an innovative deep learning framework for vessel trajectory prediction (Trans-Graph), which effectively integrates diverse features extracted from Automatic Identification System (AIS) data. The trajectories of vessels are expressed by the graphs. An architecture combing the graph neural networks (GNN) with Transformer is fabricated to process and analyze maritime data. To enhance the model’s capability in learning temporal ship position features, an auxiliary training task is implemented by using the randomly masked contextual information passing through bi-directional LSTM. The framework’s performance was rigorously evaluated using the AIS dataset from Copenhagen Port, Denmark, demonstrating significant advantages over existing baseline models in terms of prediction accuracy.

This paper proposes an event-triggered tracking control scheme of the nonstrict-feedback nonlinear systems working in both sensor-to-controller (SC) and controller-to-actuator (CA) channels. Aiming at...