Xiaoqing Zhu

Lateral stabilization of a single wheel robot applying electromagnetic force

The lateral stabilization of single wheel robot is a challenge for researchers. We proposed a new mechanism to solve the problem by applying electromagnetic force. The mechanism was described briefly, and after dynamic analyses of the robot, its state space equation was obtained. The prototype was built in ADAMS, and integrated with MATLAB a PID controller and a LQR controller were performed respectively. A simple physical experiment was carried out on inverted pendulum applying Lorentz force. Simulation and physical experiment results show that the proposed mechanism is feasible and more energy efficient.

Active disturbance rejection control of Single wheel robot

With only one contact point with ground, the balance control problem of Single wheel robot (SWR) is still a big challenge for researchers. In this paper, a new nonlinear control algorithm-ADRC (active disturbance rejection control technique) was proposed. The integration simulation by ADAMS and MATLAB of flywheel balanced SWR is presented. The integrated platform provides an avenue for developing a mechatronics system without the need of building mathematical model. The simulation results verified the robustness of the proposed ADRC method on the control of SWR.

Dynamic Modeling and Balancing Control of a Cubical Robot Balancing on Its Corner

Aiming at the problem of balancing control of a cubical robot, this paper makes a research on the balancing control of a cubical robot balancing on its corner. Using the physical prototype of cubical robot we designed as the research object, the dynamic model is derived with Lagrangian method. A balancing controller is proposed base on a nonlinear control method - active disturbance rejection control (ADRC). There are three ADRC controllers in pitch, roll and yaw direction. The internal and external disturbance of the system are regarded as the total disturbances of the ADRC controller. The effectiveness of the controller is verified in the comparison with PID algorithm with the obtained expect effect. The controller developed can provide a base for further study for balancing control of a physical prototype cubical robot.

A Cognitive Map Learning Model Based on Hippocampal Place Cells

Aiming at the environment cognition and navigation problem of autonomous mobile robots in unknown environment, a cognitive map learning model is proposed based on hippocampal place cells, which can memorize and map surroundings. The model uses the self-organizing feature map as the basic structure. Each hippocampal place cell is represented by a neural node. The robot builds up the hippocampal place cells layer through environment exploration. The simulation results show that the model has self-learning ability, which enables robots to acquire environment knowledge and establish a complete cognitive map gradually like human beings and animals, making the robots environment cognition and navigation process become more bionic and intelligent.

A RGB-D SLAM Algorithm Combining ORB Features and BOW

With1 the rapid development of robots and UAVs, navigation has become a very important topic. The combination of GPS+IMU is undoubtedly the standard of outdoor navigation. Indoor navigation, simultaneous localization and mapping (SLAM) is considered to be the key to autonomous navigation of robots in unknown environments. The traditional SLAM algorithm based on RGB-D camera adopts SIFT feature and needs GPU acceleration to overcome the drawback of SIFT feature extraction. In view of the above problems, a SLAM algorithm combining ORB features with closed loop detection is proposed. At the front end, using improved ORB algorithm to speed up the image feature point extraction, combined RANSAC with ICP to calculate the position and posture of the camera. In the back end, closed loop detection algorithm is proposed to eliminate the cumulative error of the robot, the pose of the camera are optimized by using graph optimization tool to obtain globally consistent camera pose and point cloud.

Single-wheel robot modelling using natural orthogonal complement

The modelling on SWR is a foundation work for researchers to carry out further study, thus is very important. The early modelling on SWR are mostly done by the Euler-Lagrange equations, and the Euler angles are used to calculate the attitude of SWR. The method of Euler-Lagrange equations has to calculate a lot of quadratic terms and partial differential terms, while the Euler angles are known to be prone to singularities. This paper proposed a novel modelling on SWR by using nature orthogonal complement, which is simpler and more efficient in computation than Euler-Lagrange equations. Further, the quaternions was employed to calculate the attitude of SWR, thus free of singularity problem. Moreover, the radius of the toroidal wheel has been taken into account on the model firstly, so the mathematical model is more close to the real physical situation. The dynamic modelling results have been verified by numerous experiments.

Kinematic parameter calibration of two-wheeled robot

The kinematic parameters of the two-wheeled robot has a great influence on the robot localization and path planning. Since there exist robot systematic errors caused by unequal wheel diameter and uncertainty about the effective wheelbase, it is difficult to acquire the exact kinematic parameters. In order to reduce the errors, researchers introduced the UMBmark to make calibration of robot kinematic parameters. However, the previous studies assume that the two factors influence the robot localization independently, which is not the case in fact. This paper put forward an improved kinematic parameters calibration method based on the UMBmark. Let the robot walk along a straight path and an equilateral triangle path respectively and we can calibrate kinematic parameters by measuring the actual position errors. Then the optimal length of the straight path was given out, and it is proved that the equilateral triangle path has better adaptability than the square path has. Physical experiment has been carried out, and the result verified the effectiveness of the proposed method.

UKF parameter optimization method using BP neural network for super-mini aerial vehicles

Owing to the uncertainty in determining the parameters used in unscented transformation which is the main procedure of Unscented Kalman Filter (UKF), we propose a learning method using BP neural network to optimize them adaptively. The experiments were performed on three methods, and the results show that the proposed learning method is better than traditional UKF algorithm, and the precision has an evident increase. The UKF algorithm using BP neural network parameter optimization is effective and feasible, which avoid successfully the lower efficiency and local optimal solution problem in the traditional method.