Research On Control Of Light Weight Robot With Flexible Joints

Recently, due to the features of high load-to-weight-ratio, lightweight and low power consumption etc, flexible joint lightweight robots are widely used in a variety of mobile robots or mechanical platforms, playing a more and more important role in the areas of space exploration, military reconnaissance, counter-terrorism, defusing, as well as home service. From the viewpoint of lightweight robot applications, there exist many different tasks under changing working conditions, either working in dangerous and unknown complex environment or closely contacting with human beings. Therefore, for a safe operation and high reliability, lightweight robots need not only high-precision position control but also compliance control, such that no injury on the robot and operated object will occur when interacting with the unknown environments. Furthermore, the flexibility of the robot joint increases due to the harmonic drive and the torque sensor integrated in the robot joint, which makes the control tasks of lightweight robots to be a challenging issue and results in inevitable vibrations during the robot motion and residual vibrations upon the stopping of the motion. In order to achieve a high control performance, the vibrations must be suppressed. Therefore, Cartesian impedance control as well as vibration (including residual vibration) suppression of flexible joint robots will be investigated deeply, in addition to the study of position control problems.In this work, the hardware system of a flexible joint robot is developed, including mechanical subsystem, sensor subsystem and electronics subsystem; a brief introduction to the electrical layout of the entire system is also given. A 4-DOF lightweight robot was designed based on the idea of modular design with a variety of integrated joint sensors. As a part of the robot system, a lower-level (inner-loop) controller based on Xilinx FPGA, an upper-level (outer-loop) controller based on PCI technique, and a point-to-point-LVDS high-speed serial communication bus between the lower-level and upper-level controllers (with a communication cycle of 200us) were developed as well.The position control of a single-link flexible joint robot based on backstepping approach is realized. However, the traditional backstepping approach is sensitive to the model parameters of controlled system. Therefore, a neural network-based adaptive backstepping controller is employed, which not only overcomes the problem of the sensitivity to the model parameters, but also eliminates the need of accurate dynamic model of the flexible joints robot. Moreover, the control system is not subjected to the restrictions on the range of joint flexibility, and the link acceleration and jerk signals are not required for the control realization. Simulation and experimental results confirmed the effectiveness of the proposed control approach, and at the same time, the repeatability of positioning accuracy of end-point position and attitude is reached according to the design target.We realized Cartesian impedance controller based on backstepping approach to achieve compliance control of the lightweight robot, during which the"explosion of terms"problem associated with the traditional backstepping approach occurs. To overcome this problem, an impedance controller based on the so-called Dynamic-Surface-Control-backstepping is proposed, which introduces a first-order integral filter to estimate the derivative of the virtual control input in each step of the backstepping design process; this action not only eliminates the"explosion of terms", but also filters out the sensor noise, thus improving the dynamic performance of the system. The effectiveness of this new control approach is confirmed by the compliance behavior of the flexible joint robot working in constrained environments.Finally, vibration and residual vibration due to the joints flexibility (induced by the integrated harmonic drive and joint torque sensor) are suppressed. A joint state observer based on Luenberger approach is designed in order to reconstruct the link-side position and velocity, using motor position and joint torque as the input of the observer. As for the residual vibration upon the stopping of the motion, the time-varying input-shaping technique is introduced.In this thesis, the position control and vibration suppression as well as the residual vibration suppression are realized on the first generation 4-DOF flexible joint lightweight robot. while Cartesian impedance control is implemented on the second-generation 5-DOF flexible joint lightweight robot.