SUMMARY

Traditional mechatronic systems are built to include one or more embedded processors to add intelligence, autonomy, perception, or accuracy. The computational power of these embedded processors is typically chosen based on the ability to handle the peak resource demands. However, significant decreases in power consumption, and a commensurate increase in performance lifetime, can be obtained by coordination of the processor, algorithms, and physical systems to maximize performance while still addressing power constraints. This research addresses the development of compute-aware control systems that adjust processor characteristics and control algorithms in real time to meet changing demands. At the heart of the proposed research are anytime algorithms, which adjust their complexity in response to fluctuating computational resources. Anytime algorithms are ones that can be executed in stages, with more stages executed as CPU resources become available. In situations where CPU resources are more limited, the anytime algorithms are only partially executed, while when CPU resources are plentiful, they are executed completely. There are several strategies for developing anytime algorithms, and this research will explore appropriate techniques for different types of control methodologies. Since anytime algorithms typically involve switching of controllers, the resulting closed-loop systems are hybrid systems, for which questions of stability and performance must be studied. Furthermore, abrupt switches in control laws can give rise to transient episodes of deteriorated performance, so transient management is also of concern. Experiments and simulations will be conducted to validate the effectiveness of the anytime algorithms as well as measure the performance of the physical system in both regulation and trajectory tracking operation.