How does MPC compare with other advanced control techniques?

Model Predictive Control, is a control strategy that originated in the 70s majorly focused in the chemical processes industry. MPC employs a state-of-the-art prediction methodology to anticipate the future states of a system from it's mathematical model. The major attraction about MPC is that it is able to handle MIMO - multi-input-multi-output systems unlike usual PID controllers. MPC is a highly adaptable control approach as it can be deployed to control linear as well as non-linear systems both. MPC solves an optimization problem to generate control inputs to follow a part of the reference trajectory over a prediction horizon, but only executes the control horizon worth of the inputs generated and reiterates the optimization problem.

In my opinion, MPC allows for straightforward incorporation of constraints into the control problem. Whether it’s input saturation, state limits, or other system-specific bounds, MPC can handle them effectively. MPC naturally extends to multi-variable systems. It considers interactions between different variables, making it suitable for complex processes with interconnected dynamics.

Absolutely agree with the points mentioned. A key strength of MPC, which sets it apart from other advanced control techniques, is its unparalleled ability to adhere to constraints, whether they're direct limits on states or controls (box constraints) or more complex in nature. This aspect is particularly crucial in scenarios where the real system operates near its functional limits or in environments prone to potential collisions, such as in autonomous robotics. By embedding constraints directly into the control problem, MPC not only ensures safety and operational efficiency but also enhances predictability and reliability in dynamic and uncertain environments.

One of the main benefits of MPC is it's ability to handle linear as well as non-linear constraints. Like a true AI based control methodology, constraint handling in MPC provides for extremely flexible and safe control. Even in various non-linear systems, a non-linear MPC controller is able to account for the changing the constraints of the system which is remarkable. The feedback control aspect of MPC also adds to reasons for the exceptional control performance achieved . MPC allows us to prioritize various parameters of the cost function while solving the optimization problem by assigning weights to the said parameters which are then prioritized according to the decreasing order of their weights.

The biggest disadvantage of MPC is the exceptionally high resource demand. A system running an MPC controller to control a Non-Linear system has to invest considerable amount of resources to the online MPC optimization problem which makes it very difficult to perform other online tasks which require even lesser resources like - SLAM. Also, owing to it's predictive property, MPC requires an accurate model of the system to generate appropriate control inputs, which might not always be the case as having an accurate model of the system is not always possible, especially for Non-Linear systems. Unmodelled external disturbances, if large, can significantly disrupt the performance of the MPC controller as they are not a part of the system model.

The success of Model Predictive Control (MPC) largely depends on the mathematical model's quality. Optimal results are often achieved with 'grey-box' models, which blend physical system modeling with specific application data. This approach enhances model accuracy, aligning it closely with real-world dynamics and improving MPC's performance. Despite its computational demands, a well-crafted model can make MPC highly effective, especially in handling unpredictability and ensuring system stability.