Research On Two-Wheeled Self-Balanced Robot

As a kind of intrinsicly instable wheeled mobile robot, two-wheeled self-balanced robot (TSR) has multi-variable, non-linear, strong coupling and parameter uncertainty characteristics which make it to be an imagine platform to verify many kinds of control algorithms. TSR has flexible movement and simple structure, suitable for small space, so it has a broad prospect. TSR could perform complicate motion and manipulation tasks which the multi-wheeled robot could not do, and it is very adaptable to the great-change environment or complicated tasks, such as space exploration, battlefield scout, dangerous goods transportation etc., it may also be used in toy, education and service robot fields. The research on the TSR system has the important theory and practical significance to heighten the research level and expand the application backgrounds.Because robot's structure has important effect on the self-balanced and motion control, an even-distributed molar and fixed centroidal TSR was proposed, which has flexible movement and strong anti-jamming. Those criterions make the TSR's design rational and practical. Through kinematics analysis of TSR, the kinematics model was established. The mathematic models of TSR's kinetic energy, potential energy and dissipative energy were obtained. Then, TSR's dynamic model was established by Lagrange equation. Both kinematic and dynamic model provide a theoretical basis for the design of control strategy. The self-balanced controller was designed by adaptive algorithm. The dynamic self-balanced is realized and the anti-jamming performance is increased. Simulation results verify that the algorithm has advantages in self-balancing, anti-jamming and adjusting time.For decreasing error of position and attitude estimation from inertial sensors and improving positioning accuracy, error mechanism and drift characteristic were analyzed, and the error models were established by the Levenberg-Marquardt nonlinear least-square iterative fit method. Aiming at the dynamic and nonlinear character of different position and attitude, this paper proposed two-graded decentralized isomeric Kalman Filter algorithm. It realizes integrated navigation of TSR, improves navigation's performance, and compensates for the inertial sensor error. Thus, an optimal estimation for position and attitude is achieved. Experimental results demonstrate that the method is effective and feasible. To recognize the running state efficiently and use exact motion control algorithm, a SVM-based multisensory two-graded data fusion method was presented. The running state recognition is realized and the low classified accuracy is solved. Experimental results demonstrate that the running state could be recognized efficiently and reliably.Aiming at the anomaly events, such as the wheel-slippage, the obstacle negotiation, the bump etc., the odometry would invalidate in position and attitude estimation. The Accodometry method was proposed combined with improved Gyrodometry method to fuse the encoder data and inertial sensor data. The accurate positioning of TSR is achieved and the effect of non-systematic errors on the robot position and attitude estimation is solved. The ill-effects of gyro and accelerometer inherent drift are eliminated. The positioning accuracy is increased. The experiment verifies the effectiveness of the Accodometry method. Experimental results show that the position error falls to about one-third and the orientation error falls to one-sixth.For the effective use of limited battery power and extending the run time, several methods of reducing energy consumption were compared. According to the dynamic model of TSR, the energy consumption model was established. Because the energy function with exponential form was difficult to achieve the analytic solution, a pseudo-distributed asymptotic optimization strategy based on adaptive pseudo-parallel genetic algorithm was presented. The velocity and energy consumption were optimized. An effective and normalized combination operator was proposed to reflect diversity of population. That associates with a series of methods improve the deficiencies of premature convergence, local convergence and slow convergence, and improve the performance of genetic algorithm. Experimental results prove that the optimization algorithm is effective and fit for theTSR.Finally, a simulation and a hardware experiment systems were built. The basic motion functions of TSR were verified through the hardware experiments. Besides, many experiments were carried out. The experimental results show that TSR could satisfy dynamic and real-time demand, the structure design is rational, and the control strategies are effective.