Book excerpt: Computer vision applied to guidance of autonomous robots has been generating large interest in the research community as a natural and rich way to sense the environment and extract from it the necessary features. However, due to the robots motion, vision-based sensing becomes a real-time activity that must meet deadlines in order to support adequate control performance and avoid collisions. Unfortunately, most vision-based systems do not rely on real-time techniques and exhibit poor temporal behavior, with large variations in execution time that may lead to control performance degradation and even sensing blackout periods caused by skipped image frames. In this chapter, the referred problem is identified in the scope of the CAMBADA middle-size robotic soccer team, being developed at the University of Aveiro, Portugal. Then, a new architectural solution for the vision subsystem is presented that substantially improves its reactivity, reducing jitter and frame skipping. The proposed architecture separates the vision-based object-tracking activities in several independent processes. This separation allows, transparently and relying solely on operative system services, to avoid the blocking of higher priority processes by lower priority ones as well as to set independent activation rates, related with the dynamics of the objects being tracked and with its impact on the control performance, together with offsets that de-phase the activation instants of the processes to further reduce mutual interference. As a consequence, it becomes possible to guarantee the execution of critical activities, e.g., obstacle avoidance and privilege the execution of others that, although not critical, have greater impact on the robot performance, e.g., ball tracking. Finally, many robotic applications are deployed in open environments that are hard to characterize accurately at pre-runtime. The architecture herein proposed permits managing dynamically the resources assigned to tasks, e.g. by controlling their execution rate or priority, allowing a dynamic control of the delivered QoS. This approach permits either maximizing the utilization of system resources to achieve a best possible QoS for different load scenarios or adjusting the resource utilization according to the application instantaneous requirements, granting a higher QoS to the tasks that have higher impact on the global system performance.