Skip to main content
QUICK REVIEW

[论文解读] Neural Network Model for Path-Planning of Robotic Rover Systems

Youssef Bassil|arXiv (Cornell University)|Apr 1, 2012
Robotic Path Planning Algorithms参考文献 16被引用 11
一句话总结

本文提出一种三层前馈人工神经网络(ANN),用于在复杂行星地形中导航的机器人探测车实现自主路径规划。通过离线反向传播训练,该模型使模拟探测车能够成功避开障碍物并沿最安全路径抵达目标,展示了在障碍物密集环境中的有效实时导航能力。

ABSTRACT

Today, robotics is an auspicious and fast-growing branch of technology that involves the manufacturing, design, and maintenance of robot machines that can operate in an autonomous fashion and can be used in a wide variety of applications including space exploration, weaponry, household, and transportation. More particularly, in space applications, a common type of robots has been of widespread use in the recent years. It is called planetary rover which is a robot vehicle that moves across the surface of a planet and conducts detailed geological studies pertaining to the properties of the landing cosmic environment. However, rovers are always impeded by obstacles along the traveling path which can destabilize the rover's body and prevent it from reaching its goal destination. This paper proposes an ANN model that allows rover systems to carry out autonomous path-planning to successfully navigate through challenging planetary terrains and follow their goal location while avoiding dangerous obstacles. The proposed ANN is a multilayer network made out of three layers: an input, a hidden, and an output layer. The network is trained in offline mode using back-propagation supervised learning algorithm. A software-simulated rover was experimented and it revealed that it was able to follow the safest trajectory despite existing obstacles. As future work, the proposed ANN is to be parallelized so as to speed-up the execution time of the training process.

研究动机与目标

  • 开发一种适用于在复杂、障碍物密集的行星环境中运行的机器人探测车的自主路径规划解决方案。
  • 解决在地形障碍物存在的情况下保持稳定并成功抵达目标的挑战。
  • 实现一种神经网络模型,以实现实时决策,确保安全导航。
  • 通过在障碍物条件下对探测车行为进行软件仿真,验证模型的有效性。

提出的方法

  • 设计一个包含一个输入层、一个隐藏层和一个输出层的多层前馈神经网络。
  • 采用反向传播监督学习算法对网络进行离线训练。
  • 使用带障碍物的模拟地形数据作为输入,训练网络以预测最优路径方向。
  • 将传感器输入(例如障碍物接近程度)映射到电机控制输出(例如转向和速度指令)。
  • 在地形配置及其对应的安全路径轨迹数据集上训练网络。
  • 将训练好的网络部署于软件模拟的探测车环境中,以评估其在实时导航中的性能。

实验结果

研究问题

  • RQ1前馈神经网络能否有效学习在模拟行星地形中绕过障碍物?
  • RQ2在存在障碍物的条件下,该ANN模型在路径稳定性和抵达目标方面表现如何?
  • RQ3在仿真过程中,该ANN在生成安全、无碰撞路径方面的性能如何?
  • RQ4离线反向传播训练如何影响网络对新型地形配置的泛化能力?
  • RQ5该模型未来是否可通过并行化扩展以支持实时执行?

主要发现

  • 训练后的ANN成功使模拟探测车在充满障碍物的环境中导航而未发生碰撞。
  • 探测车始终沿最安全路径抵达目标,表现出有效的避障能力。
  • 该模型在所有测试的仿真场景中均实现了稳定可靠的路径规划性能。
  • 离线训练过程有效学习了地形输入与最优导航决策之间的映射关系。
  • 仿真结果证实了使用ANN实现行星探测车自主路径规划的可行性。
  • 未来工作将集中于并行化训练过程以缩短执行时间,表明其具备良好的可扩展性。

更好的研究,从现在开始

从阅读论文到最终审阅,大幅缩短您的研究时间。

无需绑定信用卡

本解读由 AI 生成,并经人工编辑审核。