Navigation in cluttered environments with dynamic obstacles remains a major challenge for unmanned aerial vehicles (UAVs), especially when both fast response and reliable trajectory generation are required. Traditional modular planners suffer from the latency introduced by mapping and iterative optimization, whereas end-to-end methods, despite their efficient inference, often lack explicit feasibility constraints on the generated trajectories. To address these limitations, we propose a self-supervised end-to-end UAV planning framework that combines Bernstein polynomial-based feasible trajectory generation with point-cloud-based differentiable supervision for collision-aware learning. Specifically, the perception network predicts unconstrained latent variables, which are subsequently transformed into executable trajectories by a carefully designed differentiable cascaded decoder. By leveraging the properties of Bernstein polynomials, the decoder generates trajectories that naturally satisfy velocity and acceleration constraints, thus requiring no back-end optimization. Meanwhile, the supervision module uses raw point clouds as privileged information and extracts differentiable distance and gradient cues through local quadratic surface fitting. Because local quadratic surfaces can be fitted efficiently online as the scene changes, this approach enables training in dynamic environments. Extensive simulation and real-world experiments validate the effectiveness, real-time performance, and zero-shot sim-to-real capability of the proposed method.
Unmanned aerial vehicles are increasingly required to operate in constrained and wind-disturbed environments, where precise and reliable control is essential. To address this challenge, we propose a structured risk-sensitive disturbance-compensation model predictive control (SRS-DMPC) framework. Unlike conventional DMPC or uniform risk-sensitive DMPC (RS-DMPC), SRS-DMPC allocates heterogeneous risk attitudes across dynamic channels, enabling the controller to dynamically prioritise more critical states and reallocate control authority under uncertainties. Meanwhile, an Adaptive Kalman Filter (AKF) is employed to quantify and propagate uncertainty in real time. High-fidelity Gazebo simulations show that SRS-DMPC reduces position tracking error compared with baseline DMPC and RS-DMPC, and completes the task without loss of stable flight in the tested severe-wind scenario, whereas both baselines fail. Real-world flight experiments further demonstrate that SRS-DMPC achieves the lowest position tracking RMSE, with reductions of 13.0% and 26.6% relative to RS-DMPC and DMPC, and completes the tested trajectory without collision under the external wind disturbance whereas both baselines fail. These results demonstrate the disturbance-tolerance and practical feasibility of SRS-DMPC for quadrotor control in uncertain environments. GitHub page for this work along with videos is available at https://srs-dmpc.github.io/.
Magnetic microrobots exhibit significant potential for remote-controlled tasks within confined, unstructured biomedical environments, where they can interact with soft tissue substrates and achieve locomotion through surface contact. However, in environments such as the gastrointestinal tract, dynamic tissue undulations, compliant surface deformation, and spatially varying friction lead to uncertain contact disturbances. These disturbances hinder gait propulsion and kinematic responses, which makes stable path following on undulating flexible surfaces a persistent challenge. In this paper, we propose an event-triggered hybrid guiding vector field (GVF) visual path-following control approach for magnetic bipedal microrobots on undulating flexible substrates. The method uses visual feedback to localize the robot and projects its position onto a discrete reference path, from which the cross-track error and path progression are estimated. These path-state variables are incorporated into a hybrid GVF law that combines tangential progression with normal error correction to generate a continuous desired motion direction. An event-triggered mode supervisor uses the real-time error and progression states to switch among tracking, enhanced-propulsion, path-recovery, and safe-stop modes under dynamic contact perturbations. The resulting desired motion direction is converted into direction-constrained oscillating magnetic field commands to preserve the continuity of the bipedal magnetic gait. Experimental results on a vision-guided magnetic actuation platform demonstrate that the proposed method exhibits superior cross-track error suppression compared to conventional visual servoing, achieving an average cross-track error of less than 25% of the robot’s body length. Furthermore, the approach maintains stable tracking across various substrate undulation frequencies and successfully executes free-form path following on ex vivo porcine gastric tissue, demonstrating its practical viability on unstructured, flexible biological surfaces.
Existing prescribed-time control methods mostly focus on regulation or tracking control during finite time intervals. In this paper, a novel prescribed-time composite learning tracking control scheme that ensures sustained tracking over an infinite time interval is proposed for uncertain high-order fully actuated (HOFA) systems with both tracking error constraints and output constraints. First, a prediction error reconstruction mechanism is developed by utilizing online historical data and regressor extension. Then, the prediction error and tracking error are employed to construct a novel prescribed-time composite learning adaptation law. Based on the parameter estimation, a prescribed-time composite learning tracking controller is developed for HOFA systems with parametric uncertainties, ensuring that both tracking error and parameter estimation error converge to zero within prescribed time under interval excitation (IE) condition and remain zero thereafter. Furthermore, due to the inherent coupling between the constraints on tracking error and output, existing constrained control methods for HOFA systems can only address one of the two constraint issues. To overcome this limitation, we introduce a novel dual-constraint barrier function (DCBF), allowing users to independently prescribe constraining functions for both desired control performance and safety requirements. The proposed method is validated through simulations and experiments on a robot arm to show its effectiveness.
To address the degradation of position tracking performance in ball screw feed systems caused by rigid-flexible coupling, model uncertainties, and compound disturbances, this paper proposes a fixed-time prescribed performance incremental dynamic surface control method based on an improved observer. First, a hierarchical recursive incremental model of the BSFS is established from the local incremental relationships between the motor side rotational dynamics and the worktable side translational dynamics. This model reduces the explicit dependence of the control law on model state feedback terms. Second, a nonlinear fixed-time correction term is incorporated into the conventional immersion and invariance framework, and two fixed-time immersion and invariance observers are then constructed to provide online estimation and compensation for the lumped residual disturbances in the two subsystems. Then, a fixed-time prescribed performance function based on the hyperbolic tangent function is constructed. The tracking error is constrained through an error transformation. Furthermore, a dynamic surface control method based on an adaptive fixed-time filter is designed to avoid the explosion of complexity and improve the dynamic response of the system. Finally, experiments are conducted to validate the effectiveness and advantages of the proposed method.
