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
 
 
 
 
 

71 行
2.2 KiB

using System.Collections.Generic;
using Unity.MLAgents.Actuators;
using Unity.MLAgents.Analytics;
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;
ActionSpec m_ActionSpec;
ActionBuffers m_LastActionBuffer;
private bool m_AnalyticsSent = false;
internal ICommunicator m_Communicator;
/// <summary>
/// List of actuators, only used for analytics
/// </summary>
private IList<IActuator> m_Actuators;
/// <inheritdoc />
public RemotePolicy(
ActionSpec actionSpec,
IList<IActuator> actuators,
string fullyQualifiedBehaviorName)
{
m_FullyQualifiedBehaviorName = fullyQualifiedBehaviorName;
m_Communicator = Academy.Instance.Communicator;
m_Communicator?.SubscribeBrain(m_FullyQualifiedBehaviorName, actionSpec);
m_ActionSpec = actionSpec;
m_Actuators = actuators;
}
/// <inheritdoc />
public void RequestDecision(AgentInfo info, List<ISensor> sensors)
{
if (!m_AnalyticsSent)
{
m_AnalyticsSent = true;
TrainingAnalytics.RemotePolicyInitialized(
m_FullyQualifiedBehaviorName,
sensors,
m_ActionSpec,
m_Actuators
);
}
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);
m_LastActionBuffer = actions == null ? ActionBuffers.Empty : (ActionBuffers)actions;
return ref m_LastActionBuffer;
}
public void Dispose()
{
}
}
}