Unity 机器学习代理工具包 (ML-Agents) 是一个开源项目,它使游戏和模拟能够作为训练智能代理的环境。
您最多选择25个主题 主题必须以中文或者字母或数字开头,可以包含连字符 (-),并且长度不得超过35个字符
 
 
 
 
 

58 行
1.9 KiB

using UnityEngine;
using System.Collections.Generic;
using System;
using Unity.MLAgents.Actuators;
using Unity.MLAgents.Sensors;
namespace Unity.MLAgents.Policies
{
/// <summary>
/// The Remote Policy only works when training.
/// When training your Agents, the RemotePolicy will be controlled by Python.
/// </summary>
internal class RemotePolicy : IPolicy
{
int m_AgentId;
string m_FullyQualifiedBehaviorName;
SpaceType m_SpaceType;
ActionBuffers m_LastActionBuffer;
internal ICommunicator m_Communicator;
/// <inheritdoc />
public RemotePolicy(
BrainParameters brainParameters,
string fullyQualifiedBehaviorName)
{
m_FullyQualifiedBehaviorName = fullyQualifiedBehaviorName;
m_Communicator = Academy.Instance.Communicator;
m_SpaceType = brainParameters.VectorActionSpaceType;
m_Communicator.SubscribeBrain(m_FullyQualifiedBehaviorName, brainParameters);
}
/// <inheritdoc />
public void RequestDecision(AgentInfo info, List<ISensor> sensors)
{
m_AgentId = info.episodeId;
m_Communicator?.PutObservations(m_FullyQualifiedBehaviorName, info, sensors);
}
/// <inheritdoc />
public ref readonly ActionBuffers DecideAction()
{
m_Communicator?.DecideBatch();
var actions = m_Communicator?.GetActions(m_FullyQualifiedBehaviorName, m_AgentId);
if (m_SpaceType == SpaceType.Continuous)
{
m_LastActionBuffer = new ActionBuffers(actions, Array.Empty<int>());
return ref m_LastActionBuffer;
}
m_LastActionBuffer = ActionBuffers.FromDiscreteActions(actions);
return ref m_LastActionBuffer;
}
public void Dispose()
{
}
}
}