B. Armstrong. On finding exciting trajectories for identification experiments involving systems with nonlinear dynamics. The Int. Jal of Robotics Research, 8(6):28--48, 1989.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....torque vector representing the losses in the transmissions (friction) In practice, considering an industrial arm, both the losses in the reducers and the mass properties are unknown, and an identification procedure is needed. In this field, numerous methods have been proposed in the literature [10, 11, 12]. Many of them are hard to implement because they need open loop experiments [13] or disassembly of the robot [14] Finally, a simple step bystep procedure, under PD control, has appeared to be the most efficient way to identify the dynamics of three industrial manipulators that we have used in ....

B. Armstrong. On finding exciting trajectories for identification experiments involving systems with nonlinear dynamics. The Int. Jal of Robotics Research, 8(6):28--48, 1989.

....robot is known. Experiments on the device using the algorithm exposed several factors necessary for successful implementation. These include, trajectories, friction, and sensor scaling. 4. 1 Trajectories There has been much work in finding suitable excitatory trajectories for adaptive control, [1 3]. Most of this research, however, has been with respect to linear systems. The approach we took was to use a frequency rich trajectory to excite as many modes of the system as possible. This can be achieved theoretically by applying a train of impulses or applying white noise to the system. In ....

B. Armstrong. On finding 'exciting' trajectories for identification experiments involving systems with nonlinear dynamics. IEEE International Conference on Robotics and Automation, 1987.

....and must be conserved. In these situations, we should choose a training trajectory that will get the most information out of a limited number of steps. Manually designing such trajectories is a slow process, and intuitively good trajectories often fail to sufficiently explore the state space [Armstrong, 1989]. In this paper I discuss another alternative for exploration: automatic, incremental generation of training trajectories using results from optimal experiment design. The study of optimal experiment design (OED) Fedorov, 1972] is concerned with the design of experiments that are expected to ....

....that move us through the state space, we can use the techniques of OED to design training trajectories. The intent of optimal experiment design is usually to maximize confidence in a given model, minimize parameter variances for system identification, or minimize the model s output variance. Armstrong [1989] used a form of OED to identify link masses and inertial moments of a robot arm, and found that automatically generated training trajectories provided a significant improvement over human designed trajectories. Automatic exploration strategies have been tried for neural networks (e.g. Thrun and ....

B. Armstrong. (1989) On finding exciting trajectories for identification experiments.

....model of the system is available, and learning involves finding 2 Both assumptions are reasonable in different situations; if we are attempting to learn to control a robot arm, for example, it is appropriate to assume that we know over what range we wish to control it. the proper parameters. Armstrong [1989] used a form of OED to identify link masses and inertial moments of a robot arm, and found that automatically generated training trajectories provided a significant improvement over human designed trajectories. Subrahmonia et al. 1992] successfully used experiment design to guide exploration of a ....

B. Armstrong. (1989) On finding exciting trajectories for identification experiments. Int. J. of Robotics Research, 8(6):28--48.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC