Architecting

Autonomous systems are often designed based purely on their required technical capabilities. However, the operational context is equally important. We have created a design process for autonomous systems that includes these additional requirements from the outset, allowing different designs to be evaluated for different contexts.

White drone (also know as an UAV: unmanned aerial vehicle ) with four rotor blades flying in blue sky

Project team: Bugra Alkan, Seth Bullock, Kevin Galvin and Angus Johnson

Autonomy in the real world

Autonomous systems are often discussed in idealised terms that overlook wider practicalities of the real world. Context is often considered as an afterthought that the technological solution can be adapted to.

The reality is very different: context provides requirements of an autonomous system that are as essential as those of the primary function. These additional non-functional requirements (NFRs) often result from the human factors that are present in any autonomous system, such as:

  • Human interaction requirements:For example, an autonomous delivery lorry will navigate and move itself, but its design might overlook that a human driver also provides security.
  • Human-robot teaming: Autonomous systems are usually deployed by people as part of a wider team, and the autonomous system must work and communicate with its human teammates

 

Examples of NFRs

  • Cost
  • Safety
  • Security
  • Supportability
  • Scalability
  • Usability

Technological capabilities are often emphasised in the design process over the practicalities captured by NFRs, with the result that systems work inefficiently in the real world, or even don’t work at all.

NFRs should instead be incorporated into any design process for an autonomous system to be fit for purpose. However, there has not previously been a structured approach to achieve this.

Architecting autonomous systems

To address this issue, we created a two-step design process that incorporates NFRs at an early stage of autonomous system design and allows decisions to be made based on the application context.

First, we incorporated NFRs into an axiomatic design (AD) process. This considers how customer needs produce functional requirements, which are then used to design the system. The process is iterative to allow designers to think through complex interactions.

We adapted this process by including NFRs for consideration alongside functional requirements at an early and highly formative stage of the design process. This means that the NFRs influence all design parameters from the earliest stages of system design.

Second, we used an established decision-making protocol, the fuzzy analytical hierarchy process (F AHP), to produce a rank-ordered list of how well the various NFRs are met by the autonomous system produced by the axiomatic design process.

This list could then be compared with the end-user requirements for a particular application to assess the design’s suitability for that context.

Case study: a UAV charging station

We can see this process in action through our design of an automatic charging station for unmanned aerial vehicles (UAVs). The system must operate by UAVs landing across electrodes, but any physical configuration can be used.

We used the two-step AD/F-AHP process above with the identified functional requirements (such as physical dimensions) and NFRs (such as affordability and scalability) to design and evaluate two configurations: one with simple parallel electrodes and the other with a modular hexagonal array of contacts.

The evaluation process produced a rank ordering of how well the NFRs were met by each approach. This allowed a decision to be made on the appropriate design based on the context and which NFRs each would be likely to prioritise. A hobbyist, for instance, would be likely to prioritise cost over scalability, whereas the military user is instead likely to prioritise security and reliability over affordability.

This means different designs are identified as being most appropriate, despite having the same functional purpose of charging a UAV.

Context in design

Designing complex systems takes time. Adding the requirements of context to the design of an autonomous system requires even more time and consulting more widely than just with technical experts. However, our framework addresses this in a systematic way that produces robust and reliable design outcomes that will save time and money in producing fit-for-purpose autonomous systems.

Including context in design promises autonomous systems that integrate well with human operators and teams, resulting in optimal operational effectiveness.

UAV charging station - NFR rank ordering by design

Attributes ranked in descending order

Parallel electrodes

  1. Weight & affordability
  2. Reliability
  3. Maintainability
  4. Interoperability
  5. Security
  6. Scalability
Conclusion

Best for hobbyists

Hexagonal electrode array

  1. Security
  2. Weight
  3. Reliability
  4. Interoperability
  5. Scalability
  6. Maintainability
  7. Affordability
Conclusion

Best for military user