OMG I’M SOO HAPPY IT FINALLY WALKS(in simulation) so basically a few days ago I started training with a faulty model but I locked in and spent a few hours fixing it and reuploading. I also tuned the reward gains and removed some rewards I felt were too complex. I took inspiration from the setup for the learning on the Unitree Go2 and their reward system and it seems to learn super fast, walking behavior exhibted by iteration 200-300, I’ll train the model overnight to see what it comes up with. I’m super happy how this turned out. full in depth devlog will be posted later when I have time.