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
 
 
 
 
 

139 行
3.7 KiB

using System.Collections.Generic;
using System;
using System.Collections;
using Unity.MLAgents.Sensors;
namespace Unity.MLAgents.Policies
{
/// <summary>
/// The Heuristic Policy uses a hards coded Heuristic method
/// to take decisions each time the RequestDecision method is
/// called.
/// </summary>
internal class HeuristicPolicy : IPolicy
{
public delegate void ActionGenerator(float[] actionsOut);
ActionGenerator m_Heuristic;
float[] m_LastDecision;
bool m_Done;
bool m_DecisionRequested;
ObservationWriter m_ObservationWriter = new ObservationWriter();
NullList m_NullList = new NullList();
/// <inheritdoc />
public HeuristicPolicy(ActionGenerator heuristic, int numActions)
{
m_Heuristic = heuristic;
m_LastDecision = new float[numActions];
}
/// <inheritdoc />
public void RequestDecision(AgentInfo info, List<ISensor> sensors)
{
StepSensors(sensors);
m_Done = info.done;
m_DecisionRequested = true;
}
/// <inheritdoc />
public float[] DecideAction()
{
if (!m_Done && m_DecisionRequested)
{
m_Heuristic.Invoke(m_LastDecision);
}
m_DecisionRequested = false;
return m_LastDecision;
}
public void Dispose()
{
}
/// <summary>
/// Trivial implementation of the IList interface that does nothing.
/// This is only used for "writing" observations that we will discard.
/// </summary>
class NullList : IList<float>
{
public IEnumerator<float> GetEnumerator()
{
throw new NotImplementedException();
}
IEnumerator IEnumerable.GetEnumerator()
{
return GetEnumerator();
}
public void Add(float item)
{
}
public void Clear()
{
}
public bool Contains(float item)
{
return false;
}
public void CopyTo(float[] array, int arrayIndex)
{
throw new NotImplementedException();
}
public bool Remove(float item)
{
return false;
}
public int Count { get; }
public bool IsReadOnly { get; }
public int IndexOf(float item)
{
return -1;
}
public void Insert(int index, float item)
{
}
public void RemoveAt(int index)
{
}
public float this[int index]
{
get { return 0.0f; }
set { }
}
}
/// <summary>
/// Run ISensor.Write or ISensor.GetCompressedObservation for each sensor
/// The output is currently unused, but this makes the sensor usage consistent
/// between training and inference.
/// </summary>
/// <param name="sensors"></param>
void StepSensors(List<ISensor> sensors)
{
foreach (var sensor in sensors)
{
if (sensor.GetCompressionType() == SensorCompressionType.None)
{
m_ObservationWriter.SetTarget(m_NullList, sensor.GetObservationShape(), 0);
sensor.Write(m_ObservationWriter);
}
else
{
sensor.GetCompressedObservation();
}
}
}
}
}