Basic characteristics of embedded systems are that distributed,
networked electronic control units (ECUs) are integrated into their
technical or natural environment, linked to sensors and actuators with
properties like cognition and situation awareness and controlled by
robust algorithms which implies some 'intelligence' (smart systems). Applications are to be found in many domains: safety-critical
applications include aerospace, railways, automotive applications,
machinery and medical systems, as well as entertainment. Challenges relate to the dependability (safety, reliability and
security) of hardware, software and systems design, communication
technologies, sensors, actors, materials etc.
The next step on from a collection of "intelligent embedded functions"
is the development of autonomous systems. These are able to perform
complex missions in an autonomous manner, coping with unexpected
incidents and interactions from the environment.
These autonomous systems usually include machine learning system (neural network), which enables machine to learn from data gradually, instead of having everything programmed in advance. The problem is that large neural networks are bit like a black box and it is not always clear what features they learn in process. In safety critical embedded systems some types of unexpected behaviors might lead to disaster.