Boost Performance over Your Baseline with Pathmind AI

AnyLogic is proud to work with Pathmind to help you easily train AI agents to make better decisions. AI agents train quickly using a simple web application, without the need to maintain complex tooling or infrastructure. You don’t even need to think about which cloud to use. By combining the latest deep reinforcement learning methods with familiar tools, Pathmind makes it easy to adapt AnyLogic simulations for AI, experiment, and deploy Pathmind Policies to make decisions. Pathmind helps you quickly and easily discover strategies that improve your results.

Deep reinforcement learning is a type of artificial intelligence that allows an AI agent to explore a complex environment such as a simulation. Through trial and error, the agent discovers the actions and decisions that lead to the best outcomes. To guide the AI agent, domain experts and simulation modelers define what their preferred outcomes are.

Traditional solvers and optimizers, which can serve as a simulation modeler’s baseline solution, are sometimes not effective enough on models with many agents and data variability. When data shifts unexpectedly (for example, due to a shock in demand, or the introduction of new equipment), time is needed to recalculate or rewrite a solver or optimizer.

Deep reinforcement learning can work with highly variable data and makes decisions in real time, making lengthy recalculations and rewrites unnecessary. It can also coordinate the actions of many agents at once, giving simulation modelers an opportunity to optimize entirely new types of problems.

Pathmind supports simulation modelers, industrial engineers, and operations researchers in training AI agents to quickly reach their goals. Pathmind’s simple web application gives easy access to powerful deep reinforcement learning techniques and cloud computation. These techniques can beat baseline solver or optimizer solutions and discover valuable new operational strategies for business.

Pathmind Helper adds AI to AnyLogic models in just a few steps

Pathmind makes it easy to add AI to augment your simulation model's results. The Pathmind Helper lets you quickly monitor and validate the variables you want to track. It helps you set up the majority of your RL training from inside your AnyLogic model, such as the choices AI agents will face and the observations they will receive from the environment. And it lets you easily upload your AnyLogic model to the Pathmind web application, where your AI agents can be trained.

Simply export your model from AnyLogic Professional or PLE to the Pathmind web application by clicking "Export to Pathmind" within AnyLogic's Reinforcement Learning Experiment. Once there, you can run experiments using different reward functions, observation variables, and other parameters. After you have successfully trained an AI agent, Pathmind makes it easy to deploy that agent in AnyLogic or to embed it in operations and see the results of its decisions.

Go to Pathmind Helper

Getting Started with Pathmind

An introduction to Pathmind using a simple stochastic model. In the model, a state chart features Start and Intermediate states, as well as a final goal. The agent must learn to wait for a specified amount of time in the Intermediate state before it is allowed to reach the goal.

Get started

Example models

The AnyLogic example models below have been augmented for AI with the Pathmind Helper, and they include a trained AI agent to demonstrate the decision-making power of reinforcement learning. These models and their documentation are available by clicking the links.

  • 01

    Product Delivery

    Product Delivery model
    Manufacturers and distributors are spread across an area of Europe. The model looks to determine which manufacturer should deliver to which distributor to minimize wait times and distance traveled. A consideration in the model is that the nearest manufacturer may not be the best choice if it lacks inventory to fulfill an order.

  • 02

    Warehouse Routing

    Warehouse Routing model
    Two factories produce goods that can be delivered to one of two warehouses. Blocks and delays along each possible delivery route have an impact of profitability and efficiency. This model is set up to help figure out which warehouse should receive a delivery to maximize profit.

  • 03

    Supply Chain Optimization

    Supply Chain Optimization model
    A simple supply chain consists of a retailer, a wholesaler, and a factory. Keeping too little inventory at a location results in high customer wait times. Too much inventory, cuts into profits due to holding costs. This supply chain model determines the optimal inventory levels at each location to keep a balance of happy customers and high profits.