masato-ka / airc-rl-agent

AI RC Car Agent that using deep reinforcement learning on Jetson Nano
MIT License
86 stars 24 forks source link

Seek Help. The status won't change for a long time in the learning process #49

Open kavanagh-cw opened 1 year ago

kavanagh-cw commented 1 year ago

@masato-ka Thank you for your significant work that inspire me in RL After I installed it, I face some troubles during training the car. When I start the learning process via using the gamepad, the car works fine, also the status get change in the Validation box. However, after few episodes are captured, the status doesn't change anymore. I have tried both user_interface.ipynb and user_interface_without_gamepad.ipynb, but I faced the same problem. The total episodes which I captured manually don't more than 50 for each trial.

Besides, I also tried to use jetracer-auto to train the model After captured around 300 times, the status also didn't change anymore The picture below is the screenshot Screenshot (70)

I am using Waveshare Jetracer Could you give me some advice? Thank you very much