← Back to Products

Product

AUTO-PILOT

Put the controls on autopilot.

AUTO-PILOT hero image
AUTO-PILOT platform

Overview

Agent-based PID Control

AUTO-PILOT is an agent-based framework for autonomous PID tuning. By training a reinforcement learning agent to interact with a process simulator, AUTO-PILOT learns to adjust PID parameters in real-time to optimize control performance.

AUTO-PILOT features

Capabilities

From Units to Full Plants

In this software, we connect to a demethanizer column that is simulated in Aspen HYSYS. However, the framework is flexible and can be applied to any process simulator, unit operation, or full plant.

To train a surrogate model, data must be collected for different combinations of PID parameters, set points, and disturbances.

AUTO-PILOT education

Customization

Specific Control Policies Can Be Learned

Control agents can be trained to have different performance objectives, such as:

  • Rapid response with minimal rise time while maintaining acceptable stability and overshoot.
  • Slower, smoother system response that prioritizes stability and minimizes abrupt changes in output.
  • Tight control around the setpoint with minimal steady-state error and reduced oscillations.
  • Using the Integral of Squared Error (ISE) criterion to minimize the accumulated squared tracking error over time.
  • And more. You can customize the objectives!
AUTO-PILOT features

Software

Easy and Intuitive GUI

AUTO-PILOT features an easy-to-use graphical interface that allows users to set up simulations, build surrogate models, train agents, and deploy agents in real time without needing to write code.