Imagine trying to keep a car perfectly centered in its lane. Every time the car drifts even a millimeter to the left of the center line, you violently yank the steering wheel to the right. The moment it crosses the line to the right, you yank it back to the left.
In a mathematically ideal world where you can react infinitely fast and your steering wheel can move instantaneously, the car would track the center line perfectly. But in the real world, your reaction time and the physical limitations of the steering mechanism mean the car will violently oscillate back and forth across the lane.
In control theory, this violent, high-frequency oscillation is known as chattering. It is the primary pathological byproduct of Sliding Mode Control (SMC) and other discontinuous robust control strategies. While SMC is highly celebrated for its ability to keep systems stable despite massive external disturbances, chattering often makes it impractical to deploy on real physical hardware.
The Root Cause: Ideal Math meets Physical Constraints
To understand why chattering happens, we have to look at the core mechanism of Sliding Mode Control. SMC works by defining a "sliding surface"—a mathematical state where the system's tracking error is zero. The controller's job is to force the system's state trajectory toward this surface and keep it there.
To achieve absolute robustness against unpredictable disturbances, traditional SMC uses a discontinuous switching function, typically the signum function ($\operatorname{sgn}(s)$), which outputs $+1$ if the system is on one side of the surface and $-1$ if it is on the other. The control law looks something like this:
$$u = -K \operatorname{sgn}(s)$$
Where $u$ is the control input and $K$ is a large gain chosen to overcome the worst-case disturbance.
In theory, the control signal switches at an infinite frequency, forcing the system to slide smoothly along the surface $s=0$. In practice, however, two real-world constraints shatter this ideal behavior:
- Discrete-Time Sampling: As Zhian Kuang et al. point out in Feedback-based Digital Higher-order Terminal Sliding Mode for 6-DOF Industrial Manipulators, implementing continuous-time control laws on digital microprocessors introduces a sampling delay. The controller cannot switch instantly; it must wait for the next clock cycle. By the time it switches, the system has already overshot the sliding surface.
- Actuator Dynamics: Real actuators (like electric motors, hydraulic valves, or chemical pumps) cannot change state instantaneously. They possess inertia and time constants that lag behind the high-frequency commands.
The combination of these factors turns the ideal "sliding" motion into a high-frequency, finite-amplitude zig-zagging motion around the target state.
Ideal Sliding Mode (Infinite Frequency)
----------------- Sliding Surface (s = 0) -----------------
Real-World Chattering (Finite Frequency Delay)
/\ /\ /\ /\ /\ /\ /\ /\
--/--\--/--\--/--\--/--\--/--\--/--\--/--\--/--\-- Sliding Surface (s = 0)
\/ \/ \/ \/ \/ \/ \/ \/ \/
The Destructive Cost of Chattering
Chattering is not just an aesthetic issue; it is mechanically and electrically destructive.
As Fei Chao et al. observe in An Improved Fuzzy Brain Emotional Learning Model Network Controller for Humanoid Robots, this high-frequency switching causes severe mechanical wear and tear on robotic motors, leading to jerky movements and premature hardware failure. In power electronics, En‐Chih Chang et al. note in Parallel-Connected Inverter Control Method in Material Micromachining that chattering causes excessive thermal losses in power transistors and introduces unwanted electrical noise. Furthermore, in ultra-precision applications like semiconductor manufacturing, traditional SMC chattering completely degrades positioning accuracy, as highlighted by Zhian Kuang et al. in Precise Motion Control of Wafer Stages via Adaptive Neural Network and Fractional-Order Super-Twisting Algorithm.
To make robust control viable, researchers have developed several distinct paradigms to eliminate or mitigate chattering.
Strategy 1: Smoothing the Discontinuity
The most intuitive way to stop a system from violently switching is to replace the hard discontinuous $\operatorname{sgn}(s)$ function with a smooth continuous approximation, such as a hyperbolic tangent or a sigmoid function.
En‐Chih Chang et al. leverage this approach by implementing a Sigmoid Sliding Mode Function. Instead of instantly jumping between $-1$ and $+1$, the control signal transitions smoothly within a narrow boundary layer around the sliding surface. Inside this boundary layer, the controller acts like a high-gain linear controller rather than a bang-bang switch.
The Trade-off: While smoothing eliminates chattering, it sacrifices the absolute invariance property of SMC. Inside the boundary layer, disturbances can now nudge the system off target, introducing a small steady-state tracking error.
Strategy 2: Advanced Reaching Laws
Another approach is to redesign the reaching law—the mathematical rule that dictates how the system approaches the sliding surface. Traditional reaching laws force the system to crash into the sliding surface at a constant, high velocity, which guarantees a violent overshoot.
Modern strategies dynamically scale the reaching velocity based on the system's distance to the surface. For instance, in Sliding-Mode Control With Multipower Approaching Law for DC-Link Voltage of Z-Source Photovoltaic Inverters, Tan Rui et al. propose a Multipower Reaching Law (MPAL). By using an adaptive, multi-segmented power law, the controller forces the system to rush toward the sliding surface when it is far away, but smoothly decelerate as it draws close, easing onto the surface rather than smashing into it.
Similarly, Mingyuan Hu et al. introduce a New Adaptive Reaching Law (NARL) in Research on Speed Control of Permanent Magnet Synchronous Motor Based on Improved Fast Terminal Sliding Mode with Adaptive Control Law, which automatically tunes the switching gains in real-time to match the actual scale of the disturbances, preventing the over-engineering of control effort that triggers chattering.
Strategy 3: Higher-Order Sliding Modes (HOSM)
If the root cause of chattering is a discontinuous control signal entering the physical actuator, what if we could hide that discontinuity behind an integrator? This is the core intuition behind Higher-Order Sliding Modes (HOSM).
Instead of making the actual control input $u$ discontinuous, HOSM algorithms make the derivative of the control input ($\dot{u}$) discontinuous. For example, the widely used Super-Twisting Algorithm (STA) defines the control law as a combination of two terms:
$$u = -\alpha |s|^{1/2} \operatorname{sgn}(s) + \int -\beta \operatorname{sgn}(s) dt$$
Notice that the discontinuous $\operatorname{sgn}(s)$ term is placed inside an integral. Because integration acts as a low-pass filter, the high-frequency switching is smoothed out before it ever reaches the physical actuator. The actual control signal $u$ delivered to the motor or valve becomes a continuous, smooth curve, yet the system retains its mathematical robustness against disturbances. This technique is deployed effectively by Zhian Kuang et al. to achieve high-precision wafer stage positioning without actuator fatigue.
Strategy 4: Intelligent Hybridization and Observers
Why must the switching gain $K$ be so large in the first place? Because the controller is blind to the exact magnitude of the disturbances and must assume the worst-case scenario. If the worst-case disturbance is large, $K$ must be massive, which severely exacerbates chattering.
To solve this, modern control architectures hybridize SMC with intelligent estimators or disturbance observers. The architecture splits the control effort into two parts:
$$u = u_{\text{estimation}} + u_{\text{robust}}$$
- The Estimator: A neural network, fuzzy logic system, or Extended State Observer (ESO) attempts to model and predict the exact disturbance in real-time.
- The Robust Term: A sliding mode controller handles only the tiny residual estimation error left behind by the estimator.
Because the sliding mode controller only needs to reject the error of the estimate rather than the entire raw disturbance, the switching gain $K$ can be dialed down to a fraction of its original size. This hybridization drastically minimizes chattering while preserving the system's ultimate safety net.
Key Foundational Work
Control based on discrete-event models of continuous systems (1999) — contributed 2 concept elements to this direction.