An ASLab Research Seminar

Model-based Self-Awareness for Autonomy

Engineering Functional Self-Management

Carlos Hernández

Place: Aula Artigas, ETSII-UPM
Time: October 21, 2011 / 12:30-13:30

For systems to operate autonomously, their controllers have to cope with pervasive uncertainty: events in the dynamics of the plant that were unknown at design time. Classical control techniques work well when there are quantitative models of this uncertainty that can be used at design time. But we are now demanding control systems to operate with increasing qualitative uncertainty. Biological cognitive processes have provided useful inspiration to design controllers for that, although techniques are typically very dependent on the domain.

Notwithstanding, this comes at the cost of increasing the complexity of the controller: it is usually implemented as a set of components interacting to realise a set of functions designed at engineering time to address the objectives of the system. Unexpected events in the dynamics of the control system itself are thus a real threat for its success. These deviations from the expected behaviour of the controller may be due to signals arising in the plant, but their origin can also be a malfunction in the control system itself. Different methodologies have been proposed to address this problem (adaptative control, fault-tolerant control, autonomic computing), but they still do not provide a general solution to the problem. Therefore, engineers still play a role managing control systems when the unexpected occurs, redesigning on the fly the control system or the plant to cope with that.

Efficiency and autonomy demand moving this responsibility to the control system itself, making it fully autonomous. We have turned again towards the biological for inspiration, to find that consciousness, or self-awareness, can be related to the management of the cognitive processes taking place in our minds (the controllers for our bodies). It involves introspection and second-order representations, associated to the modelling, not only of the external world, but of the mental processes as well. Self-awareness could then work as an operative system supporting the cognitive processes, orchestrating their operation to make it more efficient and adaptive.

We propose an approach to engineer some of these capabilities by designing a control system that exploits a functional model of itself at runtime, so it can perform self-reconfiguration and hence cope with not pre-specified deviations from its objectives, independently of its origin (fault in the controller, unexpected in the environment...), as long as the necessary functions are available for implementation.

Find more about Carlos Hernández.