Train Multiple Agents for Area Coverage , how to move agents to predefined destinations

Train Multiple Agents for Area Coverage , how to move agents to predefined destinations using this :

https://in.mathworks.com/help/reinforcement-learning/ug/train-3-agents-for-area-coverage.html

Example :

I have 5 RL PPO agents. with 10 destinations. Want to train the agents to go to the destinations in shortest time.

How do i add destinations and train agents on the same.

Say there are different destinations [2,2],[11,2],[3,6]. Want Agent A to go to say one of the specified destination , same with agent B. both of them to be trained to go the destination in shortest time

Answer 1

[Jack]

Hey Nik,

The example you linked trains agents to maximize coverage, but if you want agents to move to specific predefined destinations in the shortest time, you’ll need to modify the reward function and action space.

Steps to Modify the Example

Define DestinationsStore your 10 destinations in a matrix:destinations = [2,2; 11,2; 3,6; ...]; % Add all 10 destinations

Assign Each Agent a Destination

• You can randomly assign a destination at the start of each episode.
• You can also assign dynamically based on a policy.

Modify the Reward Function

• Give a negative reward based on the distance to the target.
• Give a high reward when the agent reaches the destination.

Example:

function reward = getReward(agentPos, destination)

distance = norm(agentPos - destination);

reward = -distance; % Penalize distance to encourage shortest path

if distance < 0.5 % If agent reaches destination

reward = reward + 100;

end

end

Modify State Space

• Instead of covering an area, define states as (x, y) agent position and target (x, y).

Modify the Training Environment

• Instead of rewarding area coverage, focus on time-to-goal.
• Ensure the action space includes movements toward the destination.

Run Training

Modify the reinforcement learning setup from the MathWorks example and train using PPO or another RL algorithm.

This should help your agents learn the fastest paths to their destinations. Follow me so you can message me anytime with future MATLAB questions.