Reinforcement Learning Toolbox
Reinforcement Learning Toolbox offers an app, functions, and Simulink blocks to train policies using algorithms like DQN, PPO, SAC, and DDPG. Use trained policies for control and decision-making in robotics, autonomous systems, and resource allocation.
Model environments in MATLAB or Simulink, and represent policies with neural networks or lookup tables. Train single- or multi-agent setups, customize algorithms, tune hyperparameters, and simulate agents interactively or via code. Speed up training with parallel computing on CPUs, GPUs, clusters, or the cloud.
Import policies via ONNX from frameworks like TensorFlow and PyTorch. Generate C, C++, or CUDA code for deployment. Reference examples help you start quickly.
Reinforcement Learning Agents
Create model-free or model-based RL agents using algorithms like DQN, PPO, and SAC. Customize with templates or use the RL Agent block to integrate agents into Simulink.
Reinforcement Learning Designer App
Design, train, and simulate reinforcement learning agents interactively. Export trained agents to MATLAB for deployment and further use.
Reward Signals
Define reward signals to measure agent success. Auto-generate reward functions from control specs in Model Predictive Control Toolbox or Simulink Design Optimization.
Policy Representation
Start with suggested neural network architectures or define your own using lookup tables, Deep Learning Toolbox layers, or the Deep Network Designer app.
Reinforcement Learning Training
Train agents via environment interaction or existing data. Support single- and multi-agent training, with logging and real-time progress monitoring.
Distributed Computing
Accelerate training with multicore CPUs, clusters, or cloud using Parallel Computing Toolbox and MATLAB Parallel Server. Use GPUs to boost gradient computation and prediction.
Environment Modeling
Model environments in MATLAB and Simulink that interact with reinforcement learning agents, and connect with third-party tools.
Code Generation and Deployment
Automatically generate C/C++ and CUDA code from trained policies for embedded deployment. Use MATLAB Compiler and Production Server to deploy as standalone apps, shared libraries, and more.
Reference Examples
Design controllers and decision algorithms for robotics, automated driving, calibration, and scheduling. Use reference examples to start quickly.
