Motion design, control and implementation in robot manipulators
The dynamic performance of robots, specifically the tracking accuracy and motion duration, is influenced by both the nominal motion profile and the feedback control method employed. Three schemes are developed and experimentally tested to tackle the improvement of dynamic performance, in the absence of accurate dynamic models. Model Referenced Adaptive Controller Schemes (MRACS) can be designed to facilitate the characterisation of otherwise complex system dynamics. In one scheme an MRACS is used to force the robot to behave as if it were linear and decoupled, enabling simple model based dynamic tuning methods to be applied to the motion laws. Its promise as a technique is demonstrated, but the controller performance is found to be degraded by practical limitations. It is applied to both joint and Cartesian based motion laws. A computer controlled robot contains all the elements necessary for an autonomous self experimentation system. This feature is exploited in the derivation and implementation of two further schemes which are termed self learning. In these, the robot's trajectory is stored as a set of discrete data. Algorithms are developed for tuning this data subsequent to each run. Their use requires minimal knowledge of the dynamics, no additional transducers and little computation. The first of the self learning schemes is used to cyclically reduce the tracking errors. Once complete, the updating process can be curtailed. Errors on completion are close to the transducer resolution. The second of these schemes involves an incremental reduction in the duration of a given motion. Various parameters for detecting saturation are proposed and tested. A normalised ratio of peak to average velocity is found promising. Combining these two schemes, tuning for speed to near saturation then tuning for accuracy, provides a method for obtaining a near minimum time trajectory, with maximum possible tracking accuracy, at low cost.