In this study, an autonomous trajectory tracking control methodology for a mecanum wheeled mobile robot is proposed. Mecanum wheeled mobile robots are generally modeled using single contact force assumption. This assumption may be true theoretically, however, it doesn’t reflect reality. This assumption gets rid of slippage information so that a reference trajectory tracking task couldn’t be achieved with minimum amount of tracking errors. To perform desired trajectory tracking tasks with minimum tracking errors, a mathematical modeling structure which forms the autonomous system is proposed in this study. The structure proposed involves: (1) a variable contact force model, (2) a system model including kinematics and dynamics of the mobile robot and the model predictive control strategy, and (3) an autonomous trajectory tracking control system. A simulation environment and an experimental setup are built using the integration of Matlab/Simulink and ROS. Number of simulation and experimental studies are conducted by using different reference trajectories; square-shaped, Z-shaped and circle-shaped geometries. In order to show the performance of the system proposed, the reference trajectory tracking tasks are also conducted via classical PID control method. The details of the mathematical background of the autonomous system developed, built, tested and compared are presented in this paper.
Cable-driven endoscopic robots have great potential in narrow and tortuous endoscopic surgical cavities because of their high dexterity, strong compliance, and compact structure. However, when the flexible body moves inside a cavity, it is subjected to tissue-contact forces, strong nonlinear friction and hysteresis in the tendon-driven system, and unknown model-parameter uncertainties. Together, these factors form a highly dynamic and strongly coupled system that is difficult to model accurately. When conventional model-based or fixed-gain control strategies encounter such severe disturbances, they often produce substantial accumulated end-effector trajectory errors because the robust compensation intensity cannot be adjusted online according to disturbance variations, thereby seriously threatening surgical safety. To address this problem, an adaptive fast nonsingular fixed-time sliding mode control method is proposed. First, to address the potential singularity of terminal sliding modes and the difficulty of simultaneously ensuring convergence performance at different error scales, a fast nonsingular fixed-time sliding manifold combining fractional-power terms, a superlinear term, and a fast-compensation structure is constructed to jointly improve the dynamic response and steady-state accuracy. Second, a normalized adaptive robust compensation mechanism with a projection operator and leakage modification is constructed. It adjusts the compensation gain online without requiring the numerical value of the upper bound of the lumped disturbance, thereby avoiding a conservative fixed-gain design, while the boundedness of the adaptive estimates is rigorously proved. Third, a complete Lyapunov analysis is established for both the tracking dynamics and the adaptive dynamics. The designed controller guarantees practical fixed-time convergence of the tracking variables, with a convergence-time upper bound independent of the initial tracking conditions, while all adaptive parameters remain bounded. Finally, comparative simulations with mainstream methods, including fixed-time sliding mode control and nonsingular fast terminal sliding mode control, verify the effectiveness of the proposed method. The simulation results show that fast, smooth, and highly accurate trajectory tracking is maintained in the presence of disturbances and unknown uncertainties. Compared with nonsingular fast terminal sliding mode control and fixed-time sliding mode control, the proposed adaptive fast nonsingular fixed-time sliding mode controller reduces the overall position-tracking integral absolute error in the X, Y, and Z directions by 82.85% and 62.73%, respectively, and reduces the mean RMSE by 63.3% and 44.9%, respectively. In particular, the Y-direction RMSE is reduced by 91.2% relative to NFTSMC, thereby improving the robustness and control reliability of cable-driven endoscopic robots in complex luminal environments.
To address the challenges of low-quality planning paths and poor dynamic obstacle avoidance performance for Autonomous Underwater Vehicles (AUVs) in 3D underwater environments, a dynamic obstacle avoidance path planning method called GD-SAC is proposed, which improves the Soft Actor-Critic (SAC) algorithm by integrating Graph Attention Network (GAT) and the Dynamic Window Approach (DWA). Firstly, GAT is utilized to model the spatial relationships among the AUV, surrounding obstacles, and the target endpoint, thereby enhancing the algorithm’s ability to perceive environmental structures and make reasonable navigation decisions. Subsequently, the dynamic velocity window mechanism is incorporated to constrain the action space of the SAC algorithm, ensuring that the selected actions comply with kinematic constraints. To resolve the issue of sparse rewards in SAC, the reward function is improved by incorporating evaluation metrics for distance, speed, and heading derived from DWA. Additionally, priority experience sampling is introduced into the SAC algorithm to mitigate the slow and unstable training caused by random uniform sampling. The experiments demonstrate that the paths generated by the GD-SAC algorithm can effectively avoid various obstacles with higher convergence efficiency and average returns, reducing the cost of obstacle avoidance for AUVs in 3D underwater environments.
Large language models (LLMs) have shown strong capabilities in semantic understanding and reasoning, but reliably grounding their outputs into executable physical actions remains challenging, particularly on compact robots with limited onboard computation. This paper presents a modular, training-free interface in which a cloud LLM performs semantic instruction translation and task-level selection, whereas a local parser, validation gate, geometry controller, and vendor SDK execute bounded motion primitives. The contribution is a system-engineering study rather than a new low-level control algorithm: vendor-provided locomotion, cloud speech services, and an off-the-shelf open-vocabulary detector are explicitly separated from the proposed command contract, rejection policy, and observe–act–verify orchestration. Real-robot experiments cover basic and composite commands, object description, search, tracking, and height-constrained traversal. The voice-to-feedback interaction latency was 17.47 ± 2.72 s (95% CI: 16.20–18.74 s; n=20); this is interaction latency, not the local control-loop period. Across three 100-trial composite-command conditions, success ranges from 91% to 94%; across three 100-trial vision-guided tasks, success ranges from 85% to 89%. In the matched 100-scenario comparison, the full structured pipeline achieved 90% task success versus 78% for the strongest non-LLM detector-plus-controller baseline, a paired improvement of 12.0 percentage points (95% paired-bootstrap CI: 5.0–20.0; exact McNemar p=0.0042, Holm-adjusted p=0.0042). These results establish feasibility and expose current latency and perception limitations, but do not claim superiority over trained VLA policies.
Adequate localization in mobile robotics is essential for any task to be accomplished effectively and safely, particularly in changing environments. Modern localization systems are generally reliable, yet their accuracy remains conditional on scene structure and environmental stability. In feature-poor or dynamically changing environments, failures still occur and can critically impact downstream tasks. We present a framework to predict localization risk from environmental and task cues, enabling systems to anticipate and mitigate failure. For that, we introduce a map-based representation, a so-called Localization Risk Map (LRM), that spatially quantifies the level of available features as well as the level of dynamic changes by relying on state-of-the-art measures; after statistically validating them as suitable predictors of localization error, we define a Bayesian network-based model and perform parameter learning on it. By relying on a LRM and other variables, this model is able to estimate the level of localization error that might be obtained along a proposed path. We validate our method through simulated and real-world experiments with different mobile robots. The results demonstrate that our proposal consistently achieves a reliable prediction of localization error, taking into account the distribution of geometric features, the expected level of dynamic changes, or both. Also, we deploy our method in a motion planning use case, showing that the localization risk maps aid in finding the paths with the lowest localization error.
