Yefeng Yang

In complex environments, available moving space is often non-convex due to the presence of irregular obstacles, which complicates the mapping module’s ability to represent the environment with geometric symbols. This complexity hampers the effectiveness of the planning module, leading to increased computational and storage demands. To address these challenges posed by non-convex spaces, we propose an adaptive clustering convexification mapping-planning system (ACCMPS) for unmanned aerial vehicles (UAVs) navigating in unknown environments. Our approach introduces an adaptive space-searching algorithm that transforms non-convex areas into relatively convex sub-spaces represented by clusters through purposeful sampling. ACCMPS utilizes a sample-friendly 3D grid map for local mapping and constructs a storage-efficient cluster-based weighted graph for global representation, linking convex sub-spaces represented by clusters. This integration streamlines the mapping and path planning processes, enabling direct initial path generation from the weighted graph, rather than relying on traditional algorithms that blindly explore non-convex spaces. To optimize paths, we employ a particle filter on the initial graph-derived path to handle non-convexity, followed by optimization into a smooth Bspline trajectory via cost function minimization within relatively convex regions. A series of real-world experiments conducted in both narrow and expansive environments demonstrate the advantages of ACCMPS in terms of computational efficiency, storage consumption, and path quality compared to existing state-of-the-art methods.

In this paper, we propose an improved quadrotor controller by integrating the fixed-time disturbance observer (FTDO), the singularity-free recursive fast nonsingular terminal sliding mode control (RFNTSMC), and a deep reinforcement learning (DRL)-based hyperparameter optimizer. First, we decompose the overall quadrotor system into two loops: the attitude loop and the position loop. For each loop, we use an FTDO to provide a precise estimation of the disturbances within a fixed time. Then, we design for each loop an RFNTSMC to achieve finite-time tracking of a class of reference trajectories. Moreover, by adopting a DRL-based parameter optimization framework, the design parameter of our controller can be optimized while ensuring the trajectory of the closed-loop system is uniformly ultimately bounded during the learning process. We use both numerical simulations and physical experiments to illustrate the effectiveness and superiority of our results.

Traditional path-planning algorithms for robotic manipulators typically focus on end-effector planning, often neglecting complete collision avoidance for the entire manipulator. Additionally, many existing approaches suffer from high time complexity and are easily trapped in local extremes. To address these challenges, this paper proposes a goal-biased bidirectional artificial potential field-based rapidly-exploring random tree* (GBAPF-RRT*) algorithm, which enhances both target guidance and obstacle avoidance capabilities of the manipulator. Firstly, we utilize a Gaussian distribution to add heuristic guidance into the exploration of the robotic manipulator, thereby accelerating the search speed of the RRT*. Then, we combine the modified repulsion function to prevent the random tree from trapping in a local extreme. Finally, sufficient numerical simulations and physical experiments are conducted in the joint space to verify the effectiveness and superiority of the proposed algorithm. Comparative results indicate that our proposed method achieves a faster search speed and a shorter path in complex planning scenarios.

Simulating low- or zero-gravity environments on the ground is an important technology in the field of space exploration. Suspension-based gravity offloading (SGO) systems are commonly used ground-based platforms for simulating space environments. Nevertheless, the control of SGO systems presents significant challenges due to external disturbances and stringent response time requirements. This study proposes a predefined-time (PdT) observer-based PdT control framework to address the SGO system control problem. To begin, a pneumatic artificial muscle-based active unloading mechanism is introduced to address the inherent underactuation problem in the SGO system. Thereafter, a PdT disturbance observer is introduced to estimate the external disturbances acting on the SGO system. Subsequently, a type of PdT controller is investigated for the SGO system. The system, including the PdT disturbance observer and the PdT controller, is proven to be PdT stable in the Lyapunov sense. Finally, sufficient numerical simulations and physical experiments are conducted to verify the superiority and effectiveness of the proposed control method.

Suspended constant force (SCF) control is a critical technology in suspended gravity offloading systems. However, inherent underactuation, unmodelled dynamics, and external disturbances can significantly degrade control performance and even compromise system stability. In this article, pneumatic artificial muscle (PAM) actuators are used as a replacement for traditional passive dampers to address the underactuation problem. Additionally, we propose a novel systematic robust output feedback model predictive control (ROFMPC) framework, which incorporates a radial basis function neural network (RBFNN)‐based model compensator, a Luenberger state estimator, and a tube model predictive controller. The RBFNN‐based model compensator compensates for unmodelled dynamics, while the Luenberger state estimator observes external disturbances. The model predictive controller then generates the optimal control sequence. Analytical results indicate that our designed SCF system encounters similar control challenges as those in antagonistic PAM (APAM). Therefore, sufficiently comprehensive numerical simulations and physical experiments are conducted on the APAM platform to verify the effectiveness of the proposed control framework. These results demonstrate that the proposed ROFMPC framework significantly improves force trajectory tracking performance for constant force control.

This paper proposes a multi-stage dueling deep Q-network (MS-DDQN) algorithm to address the high-speed aerial vehicle evasion problem. High-speed aerial vehicle pursuit and evasion are an ongoing game attracting significant research attention in the field of autonomous aerial vehicle decision making. However, traditional maneuvering methods are usually not applicable in high-speed scenarios. Independent of the aerial vehicle model, the implemented MS-DDQN-based method searches for an approximate optimal maneuvering policy by iteratively interacting with the environment. Furthermore, the multi-stage learning mechanism was introduced to improve the training data quality. Simulation experiments were conducted to compare the proposed method with several typical evasion maneuvering policies and to reveal the effectiveness and robustness of the proposed MS-DDQN algorithm.