Two-wheeled mobile robots with deformed wheels face low stability when climbing steps, and their success rate in overcoming steps is affected by the trajectory. To address these challenges, we propose an improved hybrid genetic and adaptive particle swarm optimization (HGAPSO) algorithm to optimize the deformed wheels’ trajectory for overcoming steps. HGAPSO optimizes the maximum and minimum values of the inertial weight and learning factors of the adaptive particle swarm algorithm utilizing the region-wide search capabilities of the genetic algorithm, which substantially improves the convergence speed and adaptability. Furthermore, the analysis of the motion of the deformed wheel overcoming the steps and the examination of the potential interference during the operation are used to construct a wheel’s center-of-mass route based on fifth-order Bézier curves. Comparative simulation experiments of the trajectories optimized using different optimization algorithms under the same working conditions are designed to demonstrate the efficacy of the proposed HGAPSO algorithm in optimizing the trajectory of the deformed wheel overcoming the step. Simulation experiments were conducted using the HGAPSO algorithm to optimize the trajectories of deformation wheels for overcoming steps of various sizes. These optimized trajectories were then compared to unoptimized ones. The results showed that the HGAPSO-optimized trajectories significantly improved the success rate and stability of the mobile robot in overcoming steps.

When there are dynamic obstacles in the environment, it is difficult for traditional path-generation algorithms to achieve desired obstacle-avoidance results. To solve this problem, we propose a robot navigation control method based on SAC (Soft Actor-Critic) Deep Reinforcement Learning. Firstly, we use a fast path-generation algorithm to control the robot to generate expert trajectories when the robot encounters danger as well as when it approaches a target, and we combine SAC reinforcement learning with imitation learning based on expert trajectories to improve the safety of training. Then, for the hybrid data consisting of agent data and expert data, we use an improved prioritized experience replay method to improve the learning efficiency of the policies. Finally, we introduce RNN (Recurrent Neural Network) units into the network structure of the SAC Deep Reinforcement-Learning navigation policy to improve the agent’s transfer inference ability in a new environment and obstacle-avoidance ability in dynamic environments. Through simulation and practical experiments, it is fully verified that our method has a higher training efficiency and navigation success rate compared to state-of-the-art reinforcement-learning algorithms, which further enhances the obstacle-avoidance capability of the robot system.

SLAM (simultaneous localization and mapping) is essential for accurate positioning and reasonable path planning in outdoor mobile robots. LiDAR SLAM is currently the dominant method for creating outdoor environment maps. However, the mainstream LiDAR SLAM algorithms have a single point cloud feature extraction process at the front end, and most of the loop closure detection at the back end is based on RNN (radius nearest neighbor). This results in low mapping accuracy and poor real-time performance. To solve this problem, we integrated the functions of point cloud segmentation and Scan Context loop closure detection based on the advanced LiDAR-inertial SLAM algorithm (LIO-SAM). First, we employed range images to extract ground points from raw LiDAR data, followed by the BFS (breadth-first search) algorithm to cluster non-ground points and downsample outliers. Then, we calculated the curvature to extract planar points from ground points and corner points from clustered segmented non-ground points. Finally, we used the Scan Context method for loop closure detection to improve back-end mapping speed and reduce odometry drift. Experimental validation with the KITTI dataset verified the advantages of the proposed method, and combined with Walking, Park, and other datasets comprehensively verified that the proposed method had good accuracy and real-time performance.

Robust and highly accurate localization using a camera is a challenging task when appearance varies significantly. In indoor environments, changes in illumination and object occlusion can have a significant impact on visual localization. In this paper, we propose a visual localization method based on an ellipse-ellipsoid model, combined with object-level instance topology and alignment. First, we develop a CNN-based (Convolutional Neural Network) ellipse prediction network, DEllipse-Net, which integrates depth information with RGB data to estimate the projection of ellipsoids onto images. Second, we model environments using 3D (Three-dimensional) ellipsoids, instance topology, and ellipsoid descriptors. Finally, the detected ellipses are aligned with the ellipsoids in the environment through semantic object association, and 6-DoF (Degree of Freedom) pose estimation is performed using the ellipse-ellipsoid model. In the bounding box noise experiment, DEllipse-Net demonstrates higher robustness compared to other methods, achieving the highest prediction accuracy for 11 out of 23 objects in ellipse prediction. In the localization test with 15 pixels of noise, we achieve $ATE$ (Absolute Translation Error) and $ARE$ (Absolute Rotation Error) of 0.077 m and $2.{70}^{\circ }$ in the $fr2\_desk$ sequence. Additionally, DEllipse-Net is lightweight and highly portable, with a model size of only 18.6 MB, and a single model can handle all objects. In the object-level instance topology and alignment experiment, our topology and alignment methods significantly enhance the global localization accuracy of the ellipse-ellipsoid model. In experiments involving lighting changes and occlusions, our method achieves more robust global localization compared to the classical bag-of-words based localization method and other ellipse-ellipsoid localization methods.

To enhance the intelligence of robots, constructing accurate object-level instance maps is essential. However, the diversity and clutter of objects in indoor scenes present significant challenges for instance map construction. To tackle this issue, we propose a method for constructing object-level instance maps based on RGBD images. First, we utilize the advanced visual odometer ORB-SLAM3 to estimate the poses of image frames and extract keyframes. Next, we perform semantic and geometric segmentation on the color and depth images of these keyframes, respectively, using semantic segmentation to optimize the geometric segmentation results and address inaccuracies in the target segmentation caused by small depth variations. The segmented depth images are then projected into point cloud segments, which are assigned corresponding semantic information. We integrate these point cloud segments into a global voxel map, updating each voxel’s class using color, distance constraints, and Bayesian methods to create an object-level instance map. Finally, we construct an ellipsoids scene from this map to test the robot’s localization capabilities in indoor environments using semantic information. Our experiments demonstrate that this method accurately and robustly constructs the environment, facilitating precise object-level scene segmentation. Furthermore, compared to manually labeled ellipsoidal maps, generating ellipsoidal maps from extracted objects enables accurate global localization.

For deformed wheel-based staircase-climbing robots, the accuracy of staircase step geometry perception and scene mapping are critical factors in determining whether the robot can successfully ascend the stairs and continue its task. Currently, while there are LiDAR-based algorithms that focus either on step geometry detection or scene mapping, few comprehensive algorithms exist that address both step geometry perception and scene mapping for staircases. Moreover, significant errors in step geometry estimation and low mapping accuracy can hinder the ability of deformed wheel-based mobile robots to climb stairs, negatively impacting the efficiency and success rate of task execution. To solve the above problems, we propose an effective LiDAR-Inertial-based point cloud detection method for staircases. Firstly, we preprocess the staircase point cloud, mainly using the Statistical Outlier Removal algorithm to effectively remove the outliers in the staircase scene and combine the vertical angular resolution and spatial geometric relationship of LiDAR to realize the ground segmentation in the staircase scene. Then, we perform post-processing based on the point cloud map obtained from LiDAR SLAM, extract the staircase point cloud and project and fit the staircase point cloud by Ceres optimizer, and solve the dimensional information such as depth and height of the staircase by combining with the mean filtering method. Finally, we fully validate the effectiveness of the method proposed in this paper by conducting multiple sets of SLAM and size detection experiments in real different staircase scenarios.