Many VIO pipelines exist nowadays, but none are thoroughly implemented and optimized. Most pipelines execute a series of tasks, such as feature processing, inertial preintegration, optimization and estimate update and marginalization in a fixed sequence.
Those pipelines normally run at low and fixed measurement rates with varying execution times. By parallelizing the tasks, and changing the underlying optimization schemes, the execution speed of such a pipeline could possibly be greatly enhanced.
Many VIO pipelines exist nowadays, but none are thoroughly implemented and optimized. Most pipelines execute a series of tasks, such as feature processing, inertial preintegration, optimization and estimate update and marginalization in a fixed sequence. Those pipelines normally run at low and fixed measurement rates with varying execution times. By parallelizing the tasks, and changing the underlying optimization schemes, the execution speed of such a pipeline could possibly be greatly enhanced.
This thesis should focus on finding the bottlenecks and parallelizing the VIO tasks in a first step. Once this is sufficiently achieved, the underlying optimization or filter method could be abstracted to work more efficiently.
This thesis should focus on finding the bottlenecks and parallelizing the VIO tasks in a first step. Once this is sufficiently achieved, the underlying optimization or filter method could be abstracted to work more efficiently.