Reward function

[zqkdlmu2023]

I wanted to know how to find the task-specific reward function when training VMAS with BenchMARL, and how to modify the number of training sessions. Looking forward to your reply！

[zqkdlmu2023]

How to adjust the training time? I tried to adjust max_n_frames, but the following error occurred (see figure) ![Uploading 2024-09-25 11-32-16屏幕截图.png…]()

[matteobettini]

The reward functions of VMAS tasks are in the reward function of each scenario in https://github.com/proroklab/VectorizedMultiAgentSimulator/tree/main/vmas/scenarios

The ones of the other simulators are in the other simulators

[zqkdlmu2023]

Thanks for your answer, I have found the reward function and modified it, but it seems that the modification did not take effect when training with benchmarl. In addition, I trained the passage task with the default experimental hyperparameters, but the experimental results always failed to meet expectations.（Experimental results are shown in the figure below.） Looking forward to your answer to my question, it will be very helpful for me! ![Uploading 2024-09-25 20-03-12屏幕截图.png…]()

[matteobettini]

If you want to edit vmas you have to install it locally.

[zqkdlmu2023]

If you want to edit vmas you have to install it locally.

Yes, I modified it locally. If possible, I hope you can share the experimental parameters of the passage task.This will make it easier for me to replicate your excited work.