mlagents

大家应该都是强化学习跑Unity游戏环境吧

Traceback (most recent call last):
  File "xxxx.py", line 137, in 
    next_obs, _, done, next_info = env.step(action_step) # _ 是啥, 看上一行代码
  File "xxxxminiconda3envssim2reallibsite-packagesCogenvdecoderCogEnvDecoder.py", line 92, in step
    obs, reward, done, info = self._env.step(action)
  File "xxxxminiconda3envssim2reallibsite-packagesgym_unityenvs__init__.py", line 201, in step
    self._env.step()
  File "xxxxminiconda3envssim2reallibsite-packagesmlagents_envstimers.py", line 305, in wrapped
    return func(*args, **kwargs)
  File "xxxxminiconda3envssim2reallibsite-packagesmlagents_envsenvironment.py", line 348, in step
    outputs = self._communicator.exchange(step_input, self._poll_process)
  File "xxxxminiconda3envssim2reallibsite-packagesmlagents_envsrpc_communicator.py", line 142, in exchange
    self.poll_for_timeout(poll_callback)
  File "xxxxminiconda3envssim2reallibsite-packagesmlagents_envsrpc_communicator.py", line 111, in poll_for_timeout
    poll_callback()
  File "xxxxminiconda3envssim2reallibsite-packagesmlagents_envsenvironment.py", line 418, in _poll_process
    raise UnityEnvironmentException(exc_msg)
mlagents_envs.exception.UnityEnvironmentException: Environment shut down with return code 3221225477.

上边是在我本地windows环境跑的错，下边是我在linux环境跑的报的错(华为云)

mlagents_envs.exception.UnityEnvironmentException: Environment shut down with return code -11 (SIGSEGV).

经过N天的纠结，直到我打印了每一次的action，才发现了猫腻：

报错之前，输入的动作时 NaN, NaN，所以报了这个错

github 也有人问了，不过没解决
https://github.com/Unity-Technologies/ml-agents/issues/5675