Autonomics is a new engineering discipline that specializes in the development of autonomous systems: systems that ensure certain properties independently, without human intervention. It combines computer science, control engineering, sensor technology and mechanical engineering. Use cases range from mass individualized production to smart grids and personalized cancer therapeutics.
Key Takeaways
- Autonomous systems are not a convenience function: in the pharmaceutical industry, for example, only complete production autonomy makes life-saving cancer therapies affordable for broad sections of the population.
- The classic safety certification at development time becomes worthless with autonomous systems because the system then changes itself and safety must be continuously guaranteed at runtime.
- Autonomics as a new leading discipline would focus on the core of autonomous systems, drawing on computer science, electrical engineering, mechanical engineering and law, without fully integrating these disciplines.
- Testing of autonomous systems follows the principle of agile development: every reconfiguration at runtime must be validated by regression testing before the system accepts it.
Autonomous systems need their own discipline
Computer science alone is no longer enough to build the systems of the coming years. Peter Liggesmeyer argues for the creation of a new discipline, which he calls autonomics in reference to computer science. Its core is the ability of systems to take care of certain properties on their own without outside intervention.
The historical comparison is obvious. Computer science itself emerged from mathematics or electrical engineering, depending on the location, and has replaced these as the leading discipline, just as mechanical engineering at the time of the steam engine and later electrical engineering were formative. Autonomics would be the next step in this chain.
Such a discipline would not have to carry all the components of its basic subjects. A computer scientist knows that his work is based on hardware without being able to manufacture microchips. The specialists for microelectronics remain in electrical engineering. In the same way, autonomics could concentrate on what constitutes autonomous systems and draw on computer science, electrical engineering, mechanical engineering, business administration and, in some cases, even law.
Why several disciplines need to come together
Autonomous systems are rarely pure software. They need control and regulation technology, traditionally a field of electrical engineering, plus sensors and actuators to record data and interact with the environment, and often also mechanical engineering.
This is already evident today. Many topics appear to be computer science, but on closer inspection they have a systemic character. The safety certification of a medical device has a computer science component, but also raises mechanical engineering, electrical engineering and medical questions. A reliable answer can only be found by combining these disciplines.
The constant factor across all use cases is autonomy itself: guaranteeing certain properties, carrying out optimizations and not leaving defined corridors for settings. This property is so dominant for the architecture of such systems that it is worth having a separate discipline for it.
Where autonomous systems are already needed today
Autonomous driving is the best-known example, but the weakest. Liggesmeyer classifies it as a pure convenience function. Most people can drive decently most of the time, it would just be convenient not to have to concentrate all the time.
Fields in which autonomy is not an option but a prerequisite are more convincing. In the smart grid, thousands of small energy producers, such as private households, feed energy into the grid or draw energy from it, depending on the amount of sunlight. These grids are volatile and so confusing that people can hardly control them manually. The energy balance still has to be right, so the only option is autonomous operation.
Industry 4.0 is essentially aimed at mass-individualized products: individually manufactured but cost-effective according to the principles of mass production. The much-discussed networking is just the tool, comparable to the steam engine or the assembly line of earlier industrial revolutions. The aim is to be able to act and react better, for example to compensate for a component failure with a plan B instead of shutting down the entire plant.
Cancer drugs as an example of mandatory autonomy
The most impressive example comes from the pharmaceutical industry. CAR-T and NK cell therapeutics are produced for individual patients or small patient groups, in the case of individualized preparations from the patient’s own blood. Certain cells are extracted, genetically modified and altered in such a way that they specifically attack tumor cells.
The therapy is highly effective; a single infusion bag costs around 250,000 euros today. It is manufactured under extremely hygienic conditions: qualified personnel in hygiene suits count cells under a microscope, assess cell conditions and decide on the next process steps. Because the starting material varies greatly depending on the patient’s history, the process cannot be poured into a fixed sequence of steps.
Autonomous production technology is not a convenience here, but a prerequisite for reducing costs and making the therapy more widely available. This is precisely what distinguishes mandatory autonomy from mere convenience.
Safety must emerge at runtime, not just at development time
The classic paradigm of safety-critical systems no longer works for autonomous systems. Until now, the rule was: develop the system well, convince the certifying body, switch it on and never touch it again. A dynamically adapting system is the exact opposite of this.
Activities therefore move from development time to runtime. Liggesmeyer calls this principle “X at runtime”, where X stands for any development task. With “safety at runtime”, the system has to take care of its own operational safety at runtime.
Making a safety certification of such a system that is then invalid five minutes after being switched on because the system has changed makes no sense at all.
- Peter Liggesmeyer
To do this, these systems need a valid self-image, a priori. They must not only know their current state, but also predict whether a planned change step will lead to a safe state. This prediction by the system itself is still largely a dream of the future.
Autonomy means compromises between competing goals
Autonomous systems solve an optimization problem in which properties are mutually exclusive. If there were always a best solution, each property could be optimized separately. It is not that simple.
The classic area of tension is safety versus availability. An autonomous vehicle can continue driving or stop in a given situation. If it continues driving and makes the wrong decision, this is at the expense of safety, as it may be driving in an unsafe condition. If it stops, safety gains, but availability decreases. If the vehicle is parked on the side of the road too often, users will not accept it.
Two properties can often be improved together, but then at the expense of a third, such as costs. The system always has to make these trade-offs itself. This is precisely the hard core of autonomous driving.
Trust is created through repeated testing at runtime
Testing has always been a trust-building measure: someone has tested manually and checked the box. When test decisions are transferred to the system, this trust must be built up differently. Agile processes provide the orientation.
Repeated testing plays a central role in agile development because the status is continuously updated and there is a risk of moving existing properties in an unwanted direction. In principle, an autonomous system does the same, only without human intervention.
In terms of runtime, this means that before a system accepts a reconfiguration, it runs extensive testers. Full-scale regression testing ensures that the system continues to react in the same way as before. Only then does it accept the change.
How to prepare for autonomics today
If you want to develop in this direction, you can start with established sub-disciplines. Artificial intelligence is now an established field, and there are curriculum recommendations for data science from the German Informatics Society, from further education courses to application-oriented and undergraduate degree programs.
Both of these fields are mandatory rather than optional extras. Autonomous systems obtain their information from data that needs to be processed intelligently. Those who have mastered AI and data science will cover a large part of the future autonomous core.
Existing courses on autonomous systems and robotics are an introduction, but are often too strongly focused on a single area of application, for example with a focus on the electrical engineering aspects of robots. An independent autonomics discipline, on the other hand, should retain its own core, detached from individual application areas, just as computer science has its core independent of its applications. Such a discipline does not emerge with a razor’s edge, but takes a recognizable course. So you can already position yourself accordingly.
