6results about How to "Improve navigation performance" patented technology

The application discloses a heterogeneous flight cluster inertial navigation error cooperative correction method based on pseudo observation, which is composed of multiple unmanned aerial vehicles (UAVs) and the UAVs observe each other, and comprises the following steps: S1, acquiring inertial navigation positions of the UAVs at a current time and inertial navigation position estimation noises; S2, selecting a reference anchor point, and obtaining relative distance observation, relative azimuth observation and relative pitch angle observation among the UAVs in the cluster; S3, establishing an observation equation of the heterogeneous cluster UAVs according to the observation data obtained in the step S2; S4, constructing a pseudo observation equation according to whether the observation equation is sufficient; S5, classifying according to the rank of an observation matrix, and solving the observation equation; and S6, updating the inertial navigation position, attitude and speed of each UAV. The inertial navigation position information after correction is combined with the inertial navigation information to correct and update the navigation position, attitude and speed, and better overall positioning precision of the cluster can be obtained when the available observation quantity is insufficient.

The invention provides a mobile robot navigation method and system based on a rotary imaging radar and satellite positioning, and relates to the technical field of mobile robot navigation. Synchronously acquiring surrounding environment three-dimensional point cloud data, satellite positioning data and inertial measurement data; feature information is extracted and integrated to generate positioning attitude data; distinguishing static and dynamic characteristics according to the three-dimensional point cloud data, and generating a dynamic local environment map; when the satellite positioning signal fails, dead reckoning is carried out by combining the three-dimensional point cloud data, the inertial measurement data and the positioning attitude data before failure, and the positioning attitude data is updated; and combining the new positioning attitude data and the dynamic local environment map with the target driving data to generate an initial navigation control instruction, continuously collecting a data dynamic adjustment instruction, and controlling the robot to complete navigation. The method can adapt to a complex dynamic environment and realize high-precision reliable navigation.

The application provides a polarization tailing error estimation method based on meta-learning in a high dynamic environment, and belongs to the field of bionic polarization navigation, and comprises the following steps: a polarization angle error database caused by tailing interference under different rotating speeds is established; in an offline stage, a rotating speed related network and a rotating speed independent network are trained based on a meta-learning strategy, polarization angle errors are decoupled into rotating speed related features and rotating speed independent features; in an online stage, when performing linear motion, a polarization / inertial navigation correction dynamic model three-dimensional attitude error is used; in turning motion, the rotating speed related features are taken as states, the polarization angle inverted by a carrier dynamic model is taken as a measurement, the rotating speed independent features are taken as a measurement matrix, the rotating speed related features are estimated, the rotating speed related features and the rotating speed independent features are fused, and the estimation of the polarization tailing error is completed. The application combines offline and online learning to complete adaptive estimation and compensation of the polarization tailing error under different rotating speeds in a high dynamic environment, and improves the navigation performance of a bionic polarization integrated navigation system.

The application discloses a grid mapping optimization method and system based on an adaptive control factor, and belongs to the technical field of artificial intelligence.The application comprises a semantic map establishing module, a point cloud map establishing module, a semantic database, a map gridding module and a post-rendering module.The application carries out image preprocessing after camera image data is collected, and establishes a semantic map; carries out filtering after laser radar point cloud data is collected, and establishes a point cloud map; according to the data information of the corresponding importance score Score, the distance Dis from the central observation position and the multi-material composition Mat in the semantic database in the semantic model for establishing the point cloud map, the adaptive control factor delta of each individual model in the point cloud map is determined; according to the determined adaptive control factor, the connection between spatial points is realized in the map gridding module, and a gridded map is formed; according to pixel normal interpolation calculation of a light source, the map is restored and reconstructed through a rendering module, and the authenticity and integrity of the reconstructed map are improved.

The application provides a kind of mobile robot navigation method based on deep reinforcement learning, realizes guiding to accelerate convergence before the training of deep reinforcement learning strategy, so that mobile robot can complete safe and efficient autonomous navigation only relying on on-board sensor in complex dynamic environment.The application comprises the following steps: S1, collecting data online, comprising the following steps: S11, training navigation model in simulation environment;Define the state space of mobile robot in simulation environment, and form state vector;S12, send state vector into graph attention network, obtain state representation;S13, policy network takes state representation as input, and outputs control action;The reference action of linear velocity and angular velocity is calculated by artificial potential field method;S2, during the training phase of navigation model, train critic network, policy network and graph attention network;S3, after the training of navigation model is completed, deploy navigation model on mobile robot.

This invention discloses a scene-generalizable end-to-end autonomous navigation method based on imitation-reinforcement learning, belonging to the field of robotics technology. Addressing the problems of unstable action output and susceptibility to local optima in existing robot navigation methods, this invention constructs a TD3 network architecture, continuously interacting with the environment in a Gazebo simulation to acquire experience. The network parameters are iteratively optimized and trained using an experience replay buffer. The trained neural network outputs continuous linear and angular velocity commands for the robot based on the input state, which, after scaling, drive the robot's movement, achieving obstacle avoidance and target point navigation. Target approach rewards, collision penalties, and motion process rewards are designed to guide the policy network to improve its stable navigation ability to the target point while ensuring safe obstacle avoidance. Imitation learning is introduced to optimize the robot's motion posture; the Actor network is fine-tuned by collecting expert datasets to ensure the robot maintains posture stability during movement.