LQR vs. Pole Placement: A Comparative Analysis

1. Same Structure, Different Philosophy

At first glance, both pole placement and LQR result in a controller with the exact same structure: a full-state feedback law, $u=-Kx$. The crucial difference lies not in the implementation, but in the philosophy used to calculate the gain matrix $K$.

To explore this key distinction, let's watch a section from the MATLAB Tech Talk video on LQR.

As the video highlighted, the comparison boils down to this:

• Pole Placement: The design process starts with you, the engineer, specifying the desired closed-loop dynamics directly. You choose pole locations to achieve a target settling time, damping ratio, etc. The task is then to solve for the $K$ matrix that places the poles at these exact locations. The core question is: "What should the system's response look like?"

• Linear Quadratic Regulator (LQR): The design process starts with you defining what constitutes "good behavior" through a cost function, $J$. You don't pick pole locations. Instead, you assign penalties to state deviations (via the $Q$ matrix) and control effort (via the $R$ matrix). The LQR algorithm then finds the optimal $K$ matrix that minimizes this cost. The core question is: "What trade-offs between performance and effort are acceptable?"

2. Intuition in Design: Poles vs. Weights

While pole placement seems direct, its intuition breaks down quickly as system complexity increases.

• For a simple second-order system, you know that placing poles further to the left in the complex plane results in a faster response.
• But what about a fourth-order system like the inverted pendulum, or a higher-order multivariable system? Where should you place the third and fourth poles? How does moving one pole affect the control effort required? The answers are far from obvious.

LQR offers a different, and often more intuitive, way to tune the system, especially for the multi-input, multi-output (MIMO) systems you're focusing on. Instead of thinking about abstract pole locations, you think about the physical behavior of the system.

The following video segment demonstrates this power of LQR tuning.

The video makes a powerful point: with LQR, the tuning process is directly linked to performance objectives.

• Problem: "My actuator is saturating."
  • LQR Solution: Increase the value in $R$. This tells the controller to be more conservative with control effort.
• Problem: "The pendulum angle is not stabilizing fast enough."
  • LQR Solution: Increase the corresponding diagonal element in $Q$. This tells the controller to prioritize reducing the error in that specific state.

This approach of tuning weights is generally more scalable and intuitive for complex, multivariable systems than trying to decide where to place a large number of poles.

3. Optimality and Robustness Guarantees

This is perhaps the most significant theoretical difference between the two methods.

Pole Placement provides a controller that meets your pole location criteria, but it offers no guarantee of optimality or robustness. It's possible to place poles for a very fast response, but this might result in a system that is very sensitive to small errors in your mathematical model (i.e., it has poor stability margins).

LQR, by contrast, is provably optimal for the cost function you define. More importantly from a practical standpoint, it comes with remarkable, built-in robustness guarantees. For a single-input system, an LQR controller is guaranteed to have:

• An infinite upward gain margin.
• A gain reduction margin of at least 50%.
• A phase margin of at least $\pm 60^{\circ }$.

While the guarantees are more complex for MIMO systems, LQR is renowned for producing controllers with good stability margins. This means your controller is more likely to work on the real physical system, even if your model isn't perfect. This is a major advantage that pole placement does not provide.

4. Comparison on the Rotary Inverted Pendulum

Let's bring this comparison to the system you're interested in. The following research paper provides a direct comparison of the two techniques for stabilizing a rotary arm inverted pendulum.

The paper's conclusion highlights a common finding:

• They found pole placement was "better to improve the transient response." This is because they explicitly designed it for a specific transient response by choosing dominant poles.
• They found LQR was better for "minimizing the steady state value."
• Crucially, they note: "It is always not easy to obtain the gain matrix for state feedback [pole placement] controller easily. But in the case of LQR controller the gain matrix K can be tuned easily to obtain the desired response."

This last point echoes our discussion on intuition. While you can achieve a desired transient response with LQR by carefully tuning Q and R, the process is often more straightforward and less of a guessing game than picking a full set of poles for a complex system.

Conclusion: Which to Choose?

We can summarize the comparison in the following table:

Key Takeaways:

• Both Pole Placement and LQR are state feedback techniques that calculate a gain matrix $K$, but they do so based on fundamentally different design philosophies.
• Pole Placement gives you direct control over pole locations, which is powerful for simple systems but becomes highly unintuitive for complex, multivariable systems.
• LQR's approach of tuning weighting matrices ($Q$ and $R$) is often more intuitive, scalable, and directly related to physical performance goals (e.g., "reduce position error" or "save fuel").
• LQR provides proven optimality and, most importantly, excellent robustness guarantees, making it a safer and more reliable choice when dealing with model uncertainties. For multivariable control, LQR is very often the superior method.

Preview of the Next Module:
Congratulations on completing the LQR module! So far, all the controllers we have designed have been regulators—their primary job is to take a system from a disturbed state and return it to the zero-state equilibrium point.

However, in most real applications, we want our system to follow a specific command. We don't just want the inverted pendulum to stay balanced at $x=0$; we want to command it to move to $x=0.5$ meters and stay there. In the next module, we will build upon our state-feedback foundation to tackle reference tracking and disturbance rejection, two essential topics for creating practical, high-performance control systems.