How about designing AI control to replace PID using reinforcement learning?
This solution supports the design of next-generation control systems utilizing reinforcement learning. It is an approach that automatically acquires optimal control strategies through learning for nonlinear, higher-order delay systems and multi-degree-of-freedom control targets, which are difficult to handle with conventional PID control. The agent selects actions based on rewards, allowing for the construction of controllers that maximize performance. It is compatible with model-based design using MATLAB/Simulink's Reinforcement Learning Toolbox and Python environments, enabling the utilization of existing model assets. In addition to traditional mathematical model-driven design, it achieves integration with data-driven AI.
Inquire About This Product
basic information
Learning Method: Reinforcement Learning Algorithm Used: SAC (Soft Actor Critic) Network Structure: Actor Network, Critic Network (both are multi-layer neural networks) Tools: MATLAB/Simulink Reinforcement Learning Toolbox, Python (TensorFlow/Keras, etc.) Control Target: Electric throttle valve position control Reward Design: Minimization of error to target position + control amount penalty, etc. (estimated) Learning Settings: Number of episodes: over 1,000, batch size: 32, etc. (from graphs)
Price range
Delivery Time
Applications/Examples of results
Control design using reinforcement learning can be applied to industrial actuators, electric throttle valves, electric motor control, inverter control, and more. In particular, it may achieve more flexible responses than conventional design methods for systems that include dynamic characteristics, friction, and nonlinearity. In the application example shown, autonomous positioning is achieved using an Actor-Critic network in relation to the target value of throttle opening. In the future, applications in fields such as robotics, autonomous driving, and HVAC control are expected.
catalog（1）
Download All Catalogs
Company information
Neorium Technology responds to the diverse needs of customers involved in model-based development, consistently providing high-quality technology. In particular, we have a wealth of experience in professional services (model development, contracting, consulting, technical seminars) and product services (sales of control design support software and technical support) in the fields of control, optimization, machine learning, and system modeling. If you have any issues with model-based development, please feel free to consult us. Additionally, in collaboration with our partners, we also offer support services for ADAS simulation testing in the automotive sector. Examples include road surface creation, creation of OpenDRIVE and OpenSCENARIO data, and automation through MILS/HILS.