fix llapi and gym yamato (#4709)

gym-unity/gym_unity/envs/__init__.py

 import gym
 from gym import error, spaces
 
-from mlagents_envs.base_env import BaseEnv
+from mlagents_envs.base_env import ActionTuple, BaseEnv
 from mlagents_envs.base_env import DecisionSteps, TerminalSteps
 from mlagents_envs import logging_util
 
             action = self._flattener.lookup_action(action)
 
         action = np.array(action).reshape((1, self.action_size))
-        self._env.set_actions(self.name, action)
+
+        action_tuple = ActionTuple()
+        if self.group_spec.action_spec.is_continuous():
+            action_tuple.add_continuous(action)
+        else:
+            action_tuple.add_discrete(action)
+        self._env.set_actions(self.name, action_tuple)
 
         self._env.step()
         decision_step, terminal_step = self._env.get_steps(self.name)

ml-agents/tests/yamato/scripts/run_llapi.py

 import argparse
-import numpy as np
 
 from mlagents_envs.environment import UnityEnvironment
 from mlagents_envs.side_channel.engine_configuration_channel import (
             episode_rewards = 0
             tracked_agent = -1
             while not done:
-                if group_spec.action_spec.is_continuous():
-                    action = np.random.randn(
-                        len(decision_steps), group_spec.action_spec.continuous_size
-                    )
-
-                elif group_spec.action_spec.is_discrete():
-                    branch_size = group_spec.action_spec.discrete_branches
-                    action = np.column_stack(
-                        [
-                            np.random.randint(
-                                0, branch_size[i], size=(len(decision_steps))
-                            )
-                            for i in range(len(branch_size))
-                        ]
-                    )
-                else:
-                    # Should never happen
-                    action = None
+                action_tuple = group_spec.action_spec.random_action(len(decision_steps))
-                env.set_actions(group_name, action)
+                env.set_actions(group_name, action_tuple)
                 env.step()
                 decision_steps, terminal_steps = env.get_steps(group_name)
                 done = False