Although dynamics-based control is a quite old topic and lots of companies, such as ABB, Fanuc and so on claim they have implemented it, there have little literature about how well it works in practice and how to successfully implement it. This thesis systematically studies robot dynamics modelling, identification and control to comprehensively improve the capability of controller. Sufficient experimental results provide a guide for late researchers.

Firstly, we employ geometric method to describe the dynamics in different forms, including recursive, matrix and regressive forms. We make a clear contribution to derive regressive formula which is suitable for dynamics identification and adaptive control. The accuracy of modelling has been validated by comparing with Simmechan...[ Read more ]

Although dynamics-based control is a quite old topic and lots of companies, such as ABB, Fanuc and so on claim they have implemented it, there have little literature about how well it works in practice and how to successfully implement it. This thesis systematically studies robot dynamics modelling, identification and control to comprehensively improve the capability of controller. Sufficient experimental results provide a guide for late researchers.

Firstly, we employ geometric method to describe the dynamics in different forms, including recursive, matrix and regressive forms. We make a clear contribution to derive regressive formula which is suitable for dynamics identification and adaptive control. The accuracy of modelling has been validated by comparing with Simmechanics.

Next, we identify dynamics model by experiment. We firstly find the minimal identifiable parameters set. The regrouping formula is the first time proposed using geometric method. Secondly, we optimize a identification trajectory in Fourier-Series form. Next, the modified weighted least square algorithm is adopted to estimate parameters. Finally, we do validation experiments for verification.

Thirdly, different control schemes are thoroughly simulated in Matlab and implemented via dSpace. We deeply analyse rationality and drawbacks of Traditional PID controller. It only performs well locally due to varying motor inertias at different configurations. We firstly propose to simply add feed-forward torque to current loop. Furthermore, we select computed-torque type controllers designed by feedback linearisation or passivity-based techniques. Traditional PD, Modified PD and Augmented PD controllers are put forward. We add integral terms into PD type law after considering actual experimental performance. Finally, for counteracting effects of imperfect model, we apply adaptive control to on-line modifying model. Here we contribute to a modified adaptive control law for eliminating constraint relationship between position gain K_P and velocity gain K_D. We did thorough simulation to verify the stability, following performance and convergence. For real experiments, we introduce other two commercial controllers for comparison, i.e PDFF controller and HD controller. We study the tuning performance, trajectory accuracy and global stability. For testing adaptive control, we attach 8kg load to robot end. We explore how to successfully implement adaptive control in practice with conclusions that soft loop gain and small time delay are beneficial.

Finally, we fulfil some applications based on dynamics model. One is robot teaching by dragging. The dynamics compensation provides assistance force for ease of dragging under a modified control scheme. Besides, we explore energy efficient trajectory planning for P2P movement by considering dynamics constraints. It can be simplified with B-spline parametrization of path which leads to only optimize control points.