Model Predictive Control (MPC) in Robotics

What is it?

Model Predictive Control (MPC) is an advanced control strategy that uses a model of the robot to predict its future states across a short time window, then solves an optimisation problem at every step to pick the control input that minimises a cost (such as tracking error plus energy).

How it works

At each control step, MPC: (1) predicts the trajectory of the robot for the next N steps using a dynamics model, (2) formulates an objective function (e.g., minimise position error + penalise large control efforts) plus constraints (e.g., max joint torque), (3) numerically solves this constrained optimisation, (4) applies only the first control input of the optimal sequence, then re-plans on the next step. This rolling-horizon approach is robust to disturbances.

Real-world example

Boston Dynamics Atlas uses MPC at 1 kHz to plan whole-body motions during parkour. Tesla Autopilot uses MPC for lane keeping. Quadrupeds like ANYmal and Unitree B2 use MPC to walk over rough terrain.

Why it matters for robotics

MPC handles constraints elegantly — something PID cannot. It is the dominant control method for high-performance modern robots. Any senior robotics role in India increasingly expects familiarity with MPC.

See also

• Feedforward Control
• Feedback Loop
• Pid Controller