Complete joint domain randomization, better control command tracking on complex terrain, and relatively stable training, perfect victory over Mujoco
It is recommended to test a model with around 4,000 rounds, 10,000 because 10,000 rounds were trained
完整的关节域随机化，更好的复杂地形上控制命令跟踪，比较稳定的训练，完美战胜mujoco
建议测试4000轮左右的模型，10000是因为训练了10000轮