Sunday, November 22, 2026

When Normal Functioning Is Not Enough: An Idea from 1980s Robotics

 From the Region of Rigid Fixation to the Resilience of Modern Systems

While working on my upcoming book, Everything Is Normal… Until It Isn't, I continue to revisit my early research in robotics. Occasionally, I discover ideas that take on a surprisingly new meaning almost forty years later.

Recently, I scanned several pages from Grippers of Industrial Robots (Skhvaty Promyshlennykh Robotov), a book by my doctoral advisor, I. B. Chelpanov, and S. N. Kolpashnikov, published in Leningrad in 1989.

This book is particularly meaningful to me because one of its chapters is based substantially on my dissertation research.

But this time, something else caught my attention: a concept introduced before my own work on the Area of Normal Functioning (ANF).

It was called the Region of Rigid Fixation.

How Much Can a Robot Gripper Withstand?

Imagine an industrial robot holding a workpiece.

The workpiece is subjected to various forces and moments: gravity, accelerations during robot movement, vibrations, and interactions with other mechanical components.

Under what combinations of external forces and moments will the object remain securely held?

This is precisely the question addressed by the Region of Rigid Fixation.

The external loading is represented by six components: three forces (Fx, Fy, Fz) and three moments (Mx, My, Mz).

Together, they form a six-dimensional space. Each point represents a particular combination of external forces and moments acting on the object.

Within a certain region of this space, the object remains in equilibrium without moving relative to the gripper.

At the region's boundary, limiting conditions are reached, beyond which movement may become possible.

Chelpanov and Kolpashnikov called this the Region of Rigid Fixation.

Its shape and size depend on the gripper's geometry, gripping force, contact conditions, and friction. The book even provides three-dimensional sections of this six-dimensional region, illustrating their polyhedral geometry.

An especially interesting observation is that knowledge of expected external loads makes it possible to solve the inverse design problem: selecting gripper parameters that provide sufficient holding capability without unnecessary gripping force.

This is more than an assessment of the current operating state. It is an investigation of the system's functional limits.

Two Regions, Two Perspectives

Later, in my doctoral research, I developed and applied the concept of an Area of Normal Functioning to the motion of robotic systems.

Over subsequent decades, this concept evolved through my research on computer performance, capacity management, and anomaly detection.

Today, I use the term Area of Normal Functioning (ANF) for this broader framework.

Revisiting the Region of Rigid Fixation reveals an important distinction.

ANF describes a region of system states or operating conditions. The Region of Rigid Fixation describes a region of external disturbances that a mechanical system can withstand while maintaining fixation.

The first asks:

Under what conditions is the system functioning normally?

The second asks:

What external disturbances can the system withstand without losing its required functional properties?

These are different questions, and I believe studying them together may provide a more complete understanding of system behavior.

A Computer System Can Be Normal but Fragile

Consider two servers running identical applications.

Both show 45% CPU utilization and an average response time of 100 milliseconds. Their observed performance metrics are within normal statistical boundaries.

However, additional load testing reveals a significant difference.

The first server can accommodate another 500 requests per second before response time exceeds the specified limit.

The second can handle only 150 additional requests per second.

Both servers appear perfectly normal under their current workloads.

Yet their ability to tolerate additional demand is dramatically different.

A normal operating state does not necessarily imply a healthy reserve of functional capacity.

A system may gradually lose resilience without displaying any obvious anomalies under its usual workload.

This is where the old robotics concept becomes relevant again.

Extending ANF into the Space of External Disturbances

Suppose a system is described by two sets of variables.

The first set, X, represents its internal state and performance characteristics.

The second set, D, represents external disturbances or demands.

Using historical observations and statistical methods, we can construct an ANF describing normal values and relationships within X.

But what if we also construct a region in D-space containing disturbances under which the system can continue to meet its functional requirements?

For an IT application, the disturbance variables might include request arrival rates, network delays, database latency, or other workload drivers.

We could then study two evolving regions:

  • The Area of Normal Functioning, describing observed system behavior.

  • The region of admissible external disturbances, describing what the system can withstand while satisfying specified requirements.

An important distinction must be preserved.

Statistical normality is not necessarily equivalent to acceptable performance.

For example, a dynamically updated ANF might accommodate gradually increasing response times. Yet the application may already violate its Service Level Objective (SLO).

Therefore, the admissible disturbance region should be defined with respect to independent functional requirements, not merely the observed statistical baseline.

This leads to an intriguing possibility:

The ANF may remain almost unchanged while the region of admissible disturbances gradually shrinks.

Such a change could reveal functional degradation before traditional anomaly detection identifies a problem.

Measuring the Reserve of Resilience

In my earlier work, I introduced the Exception Value (EV) metric to quantify deviations beyond statistically determined normal boundaries.

The connection with the Region of Rigid Fixation suggests another possible measurement.

Consider the distance between the current external disturbance and the boundary of the admissible disturbance region.

We might interpret this as a robustness margin.

EV measures observed departures from normal behavior.

A robustness margin could measure how close the system is to losing its ability to tolerate additional external stress.

Of course, multidimensional disturbances require an appropriate definition of distance, especially when variables have different units and physical meanings.

Nevertheless, the idea opens an interesting research direction.

Recently, my colleague Paul Doolittle offered a useful analogy while discussing ANF and Nassim Nicholas Taleb's Antifragile.

In pharmacology, an organism may develop tolerance to a drug and establish a new normal. However, adaptation does not necessarily mean that the organism has become stronger.

To evaluate actual capability, one must compare responses to the same stimulus under controlled conditions.

The same principle applies to computer systems.

Periodic standardized load tests can tell us whether a system has genuinely improved or merely adapted to changing conditions.

More broadly, systematic stress testing could help us experimentally construct and track the region of admissible external disturbances.

Rediscovering an Old Idea

The study of system resilience under external disturbances is not new.

Control theory already includes concepts such as robust invariant sets, viability kernels, and stability margins.

I am not suggesting that these mathematical ideas are original discoveries.

What I find promising is the possibility of combining a dynamically estimated ANF with an analytically or experimentally determined region of admissible disturbances.

Together, they could help answer three fundamental questions:

  1. How is the system functioning now?

  2. Does its behavior satisfy independently defined requirements?

  3. What additional disturbances can it withstand before those requirements are violated?

And perhaps most importantly: how do these capabilities change over time?

There is also a personal dimension to this discovery.

Almost forty years after my early robotics research, I find myself revisiting the work of my academic predecessors and recognizing connections between a mechanical problem—keeping a workpiece securely held by a robot gripper—and a much broader question concerning the resilience of modern technical, organizational, and even biological systems.

Perhaps normality should be studied not only in terms of what a system is doing, but also in terms of what it is capable of withstanding.

That distinction may help us recognize approaching failure while everything still appears perfectly normal.


Historical reference: I. B. Chelpanov and S. N. Kolpashnikov, Grippers of Industrial Robots (in Russian), Mashinostroenie, Leningrad, 1989, Section 8.5, pp. 179–182.

This article is part of my ongoing research for the book Everything Is Normal… Until It Isn't.

No comments:

Post a Comment