Peter Martin, Director of Research & Development at Quanser, and co-chair of the Complex Systems Toolkit Working Group, reflects on the importance of engineers understanding complex systems when working in the field of intelligent robotics.
Peter Martin, Director of Research & Development at Quanser, and co-chair of the Complex Systems Toolkit Working Group, reflects on the importance of engineers understanding complex systems when working in the field of intelligent robotics.
“In late 2024 I had the opportunity to join the EPC Complex Systems Toolkit team as co-chair of the working group. At the time I felt a little fraudulent, as the intricacies of complex systems thinking was new to me. I had brushed up against complex systems numerous times over the years as I had studied and worked in the world of robotics for over 20 years. However, I had never discovered the world of formal complex systems analysis. Looking back, this is a perfect validation for the need to create a toolkit to better prepare students for careers like mine. As I have learned more over the last 18 months about the tools and techniques that systems engineers employ to model and manage complexity, the critical value that these techniques offer engineers in the world of intelligent robotics has become obvious. As we hear often in the field of engineering lab equipment for the academic space, “I wish I’d had this when I was at university”.
The other reassuring aspect of my experience, for me, is that I’m not alone. A growing need for better approaches to managing complexity has emerged in industry over the last couple of decades as robotics and their governing systems have become increasingly integrated into society. This transition of robotics out of the structured environment of the factory floor and into direct contact with both the dynamic and unstructured world and the public, has introduced a high degree of non-linear predictability, complex interactions with multiple robotic agents, and emergent behaviours as the decision-making algorithms that dictate robotic behaviour adapt. All of these elements are central to the world of complex systems analysis.
At a high level, modern robotics systems no longer represent technical engineering challenges in the narrow, discipline-specific sense that engineers would traditionally have seen in higher education. They are complex adaptive systems that routinely demonstrate behaviours that emerge from interactions with their environment rather than being fully specified in advance. A robot navigating a hospital corridor, a swarm coordinating warehouse logistics, or a surgical assistant adjusting in real time to tissue variability represent challenges in undefined, non-linear, and largely unpredictable spaces. Students, and later robotics engineers who lack a complex systems vocabulary are essentially tasked with trying to understand emergence without the tools to describe it.
An example that I like to use is one that we encountered a couple of years ago: a team of mobile robots transporting parts around a manufacturing space. In many cases, the agents (ground robots, arms, etc.) in this scenario are programmed with independent control and decision-making code to govern their behaviour, with some overarching supervisory code to manage tasks and assignments. The ground robots would have algorithms to localise, path plan, navigate, and avoid obstacles while communicating with other complementary agents and central task management. However, as I have learned, complexity lies in the emergence of unexpected interactions between the agents and their environment. How they avoid each other and the environment while achieving their tasks is largely a complex non-linear system where conflicts can routinely delay or disrupt their operation. Introducing more sources of disruption such as humans, unstructured environments, weather conditions etc. only makes dealing with unpredictable scenarios more and more complicated using traditional techniques.
Luckily, many of the tools and techniques that are highlighted in the toolkit have direct applications to the challenges faced by engineers in the world of robotics. Causal Loop Diagrams (CLDs) are an excellent way to model the feedback dynamics that are at play in adaptive control systems. When a robot’s perception system updates its world model based on changes in what the sensors can perceive, that leads to changes in its action policy that when executed create a feedback loop. These diagrams are a great way to visualise and analyse these loops. Agent-Based Modelling (ABM) is directly relevant to the scenario I described above where swarms of robot must be coordinated or manage human-robot interaction scenarios. Using these simulation tools, engineers can test and manage emergent fleet behaviour without hardware. If things do go sideways, Fault Tree Analysis is a common approach to mapping causes and evaluating data to help develop robots that work in safety-critical applications. Finally, for long-term operations such as field robotics missions, Systems Dynamics Modelling can be a useful tool for predicting and managing a robot’s resource consumption (battery, compute, bandwidth) depending on the required task performance over time.
In addition to these considerations, there is a whole world of network modelling and the management of behaviour stemming from machine learning and applied AI algorithms that also overlaps quite closely with complex systems. Engineers that understand emergence, feedback loops, and attractors are far better equipped to reason about why a robot does something unexpected, than students who only have a component-driven technical understanding of the behaviour of an intelligent robot. Beyond the decisions, at an actual component level there are critical decisions that need to be made for efficient deployment of physical and edge AI algorithms. What data is processed locally and what goes to the cloud, when models are updated and how decision making is distributed across a robot swarm are exactly the kind of questions that systems thinking trains engineers to answer. Systems tools are ready to help, including influence diagrams to manage information exchange and action planning.
Overall, the field of complex systems introduces a set of tools, techniques, and mental models that are increasingly essential to robotics engineers that seek to prepare their agents to be effective in performing complicated tasks in increasingly complex systems.”
Intelligent robotics, and the more recent applications to physical AI, generally refers to artificial intelligence systems that are embedded in and interact directly with the physical world, rather than operating purely in digital environments. For engineering students, learning about intelligent robotics and physical AI workflows matters because it sits at the intersection of software, hardware, and real-world problem solving. It helps engineers think holistically: not just “does the code work?” but “does the system behave safely and effectively in the real world?”.
We are seeking experts in intelligent robotics, from academia, industry, and engineering organisations, to develop resources on this topic for publication in the Toolkit. These resources will inform, guide and aid educators to embed teaching on intelligent robotics into their engineering lessons, modules or courses.
This post is also available here.
Any views, thoughts, and opinions expressed herein are solely that of the author(s) and do not necessarily reflect the views, opinions, policies, or position of the Engineering Professors’ Council or the Toolkit sponsors and supporters.