Markov Property

3 min readMay 31, 2023


The Markov Property in Markov Decision Processes (MDPs) is a fundamental concept that significantly impacts the agent’s decision-making process. This property states that the agent’s decisions depend solely on the current state, disregarding the entire history of past states and actions. In essence, the current state encapsulates all the pertinent information required for making an optimal decision regarding the next action to take.

This unique property grants the agent the ability to make decisions without relying on a complete recollection of the past sequence of states and actions. Instead, it emphasizes the immediate state and utilizes the available knowledge to select an action. By doing so, the agent is not burdened with the necessity of storing and retrieving a lengthy history of past events, enabling memoryless decision-making.

The Markov Property serves as a simplifying factor in the decision-making process, as it reduces computational complexity and memory requirements. It empowers the utilization of efficient algorithms, such as dynamic programming, reinforcement learning, and Monte Carlo methods. These algorithms can estimate value functions and derive optimal policies based solely on the information provided by the current state.

By focusing exclusively on the current state, the agent can make decisions more efficiently and effectively. This approach enables the agent to adapt promptly to changes in the environment, as it is not hindered by the burden of analyzing a vast history of states and actions. Additionally, this property facilitates the agent’s learning process, as it can accumulate knowledge over time and learn to maximize its cumulative reward.

In conclusion, the Markov Property plays a crucial role in Markov Decision Processes by enabling memoryless decision-making based solely on the current state. This simplification leads to more efficient decision-making, reduced computational complexity, and the ability to adapt and learn in dynamic environments, ultimately allowing the agent to optimize its cumulative reward over time.

Let’s imagine a robot that needs to perform a task by navigating through a room. The room can be in different states: the robot’s current position, the condition of objects in the room, whether doors are open or closed, and so on. The robot utilizing the Markov Property means that it makes decisions based solely on the current state.

For example, let’s say the robot’s task is to collect scattered objects in the room. The robot has three possible positions to move to: corner, middle, and next to the door. In each state, the robot can transition to different positions with certain probabilities. Additionally, the robot has a probability of detecting an object in a given position.

At each step, the robot observes the current state and makes a decision. For instance, if the robot is in the corner and it detects an object there, it will utilize this information to collect the object and stay in the same position. However, if the robot doesn’t see an object in the corner, it has the option to transition to a different position in the next step. By solely observing the current state, the robot makes decisions and adjusts its movements accordingly.

Thanks to the Markov Property, the robot can make effective decisions without remembering or analyzing past events. It relies solely on the current state to select the most appropriate action and efficiently complete the task of object collection.

This example illustrates how the Markov Property operates in an MDP. The robot’s decisions are based solely on the current state, and it doesn’t require information from the past to make its choices.