Hello, I am training a deep reinforcement learning based model for collision avoidance. We are using a custom mobile robot model in Gazebo 11. In the current set up, the deep reinforcement learning policy starts with random weights and the model learns from the experience in Gazebo sim. Whenever the robot collides, it gets reset to the start location using /gazebo/setmodelstate service. I am able to start the training but after 1 to 2 hours into training, the mobile robot stops responding to the /cmdvel messages and the training stops. I am not sure why this is happening. If I just sent random /cmdvel commands for a very long period the robot model works fine. I assume the problem happens because of frequent resetting of the robot using the /gazebo/setmodelstate service. Has anyone faced similar problem? If yes, how did you solve this issue?