Reinforcement Learning Agents generating zero episode/average reward!

I have a multi agent rl environment. Reward functions written for each agents work perfectly but plotted episode reward is always zero. I am using 2-3 DDPG agents.

[Emmanouil Tzorakoleftherakis]

Can you share some details on your implementation? Maybe the shipping examples on multi-agent RL can help.

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

https://www.mathworks.com/help/reinforcement-learning/ug/train-agents-for-path-following.html

https://www.mathworks.com/help/reinforcement-learning/ug/train-2-agents-to-collaborate.html

If episode reward is always zero for all your agents, check the IsDone flag and make sure it's not always true.

[Sunandan Adhikary]

The isDone flags are zeroed. While the model had 2 agents, they used to work. But now on adding another agent, all of them just shows zero reward! And i followed those shipping examples to use the tool box for multiple agent.

I am just using a function block as a system and 3 ddpg agent objects, created in main file and used in simulink model. A helper block computes reward, observation matrix etc. for each agent. There are no errors in building the project, only during the training using these objects, plotted rewards are always zero.

[Ari Biswas]

Please share a reproduction model and necessary files so that we can investigate this issue. You can also create a technical support case and share your files if you dont want to share them here.

[Sunandan Adhikary]

Thank you Ari. I have created a Support Case (04597325).

[Leonardo Malburg]

@Emmanouil Tzorakoleftherakis

Great!

Is Done was the problem for me!

Thank you