未来战争对机器人的需求_战争的未来会是什么样

最新推荐文章于 2024-09-09 23:28:21 发布

weixin_26752075

最新推荐文章于 2024-09-09 23:28:21 发布

阅读量465

点赞数

文章标签： python 人工智能 java

原文链接：https://towardsdatascience.com/what-the-future-of-war-could-look-like-ca27162bdcf3

版权

未来战争对机器人的需求

内容丰富 (Informative)

介绍 (Introduction)

One of the beauties of Artificial Intelligence(AI) is its applicability across industries, and the defence industry has been dramatically impacted by advancement in AI and related areas.

人工智能(AI)的优点之一是其在各个行业中的适用性，而人工智能和相关领域的发展极大地影响了国防工业 。

F-16 fighter jets are impressive mechanical beasts traditionally flown with the skills of a human pilot.

F-16战斗机是传统上以飞行员技能飞行的令人印象深刻的机械野兽。

But in recent years there have been academic groups and labs that have explored the potential of leveraging AI for two things:

但是近年来，有一些学术团体和实验室探索了在两方面利用AI的潜力：

Replacing the human component of fighter jets with an embedded AI system.
用嵌入式AI系统取代战斗机的人为因素。
Reducing the footprint of human manoeuvring responsibilities.
减少人为操作责任的足迹。

This article will explore DARPA’s AlphaDogFights and the considerable role AI is playing within the highly awaited AlphaDogFight that took place between 18th and 20th Aug 2020.

本文将探讨 DARPA的AlphaDogFight 和AI在2020年8月18日至20日发生的备受期待的AlphaDogFight中扮演的重要角色。

什么是DARPA的AlphaDogFight？ (What Is The DARPA’s AlphaDogFight?)

Let’s start with a brief background on DARPA.

让我们从DARPA的简要背景开始。

DARPA(Defence Advanced Research Projects Agency) is a US Department of Defence agency created in 1958.

DARPA(国防高级研究计划局) 是美国国防部于1958年成立的机构。

The agency focuses on the utilization and implementation of advanced technologies for military purposes.

该机构致力于军事目的先进技术的利用和实施。

Although the DARPA primarily focuses on military tech, the agency has also impacted several widely used technologies such as networking systems and the internet.

尽管DARPA主要关注军事技术，但该机构还影响了几种广泛使用的技术，例如网络系统和互联网。

A ‘dogfight’ is a battle between two or more fighter aircraft and involves strategical manoeuvring, combat and planning.

“ 缠斗 ”是两架或两架以上战斗机之间的战斗，涉及战略机动，战斗和计划。

Dogfights were widespread in WW1, and the first scenario that involved combat between two aircraft was in Mexico in 1913.

第一次世界大战中发生了斗殴，第一次涉及到两架飞机之间的战斗是在1913年的墨西哥。

The DARPA AlphaDogFight is a series of trials that tested the capabilities of AI algorithms executing dogfight manoeuvres, combat and strategies.

DARPA AlphaDogFight 是一系列试验，测试了执行狗斗动作，战斗和策略的AI算法的功能。

These AI algorithms performance were evaluated in a highly sophisticated simulated aerial environment.

在高度复杂的模拟空中环境中评估了这些AI算法的性能。

The first trial of the series was held last year November and the second took place on January 2020.

该系列的第一次审判于去年11月举行，第二次审判于2020年1月举行。

Eight teams have taken part in the previous two trials, and the third trial of the series is the final trial that took place this week. The last trial involved each team’s AI agent controlling an F-16 fighter jet in a simulated environment.

八支球队参加了前两次试验，而系列赛的第三次试验是本周进行的最后试验。最后的试验涉及每个团队的AI代理在模拟环境中控制F-16战斗机。

First, it’s a face-off against other AI algorithms, on the second day, the team’s AI agent then face each other. The final day of the third trial saw a simulated dogfight between the top team’s AI agents and a human-controlled F-16 jet, all in a simulated environment.

首先，这是与其他AI算法的对抗，第二天，团队的AI代理便会面对面。在第三次试验的最后一天，高层团队的AI特工与人为控制的F-16战斗机之间进行了模拟的混战，所有这些都在模拟环境中进行。

目的 (Purpose)

The AlphaDogFight is held to find methodologies of developing AI systems that can shift the responsibilities of aerial combat and manoeuvring from the human pilot to the AI system.

举行AlphaDogFight是为了寻找开发AI系统的方法，这些方法可以将空战和机动的职责从人类飞行员转移到AI系统。

To achieve this grand aim, there has to be a high level of trust in the developed AI controlling system.

为了实现这一宏伟目标，必须对发达的AI控制系统高度信任。

有关AI的更多信息？ (More On The AI?)

Major industries have transformed through the advent of recent developments in AI.

随着人工智能的最新发展的出现，主要行业已经发生了转变。

Over the decade, there’s been a considerable increase in the automation of systems and vehicles, which have all been enabled through the advancement of fields such as deep learning and reinforcement learning.

在过去的十年中，系统和车辆的自动化已经有了相当大的增长，而这些都可以通过深度学习和强化学习等领域的发展来实现。

It’s no surprise that we are currently seeing more efforts to direct the promise of AI towards defensive and military programs.

不足为奇的是，我们目前正在看到更多的努力将AI的承诺引导到防御和军事计划中。

Let’s dive briefly into how AI is utilized in dogfights.

让我们简要地探讨一下如何在混战中利用AI。

Most of the team involved in the DARPA AlphaDogFight trials are developing their AI agents by employing a combination of novel machine learning techniques and knowledge-based algorithms at varying levels.

参与DARPA AlphaDogFight试验的大多数团队正在通过结合新颖的机器学习技术和基于知识的算法在不同级别上开发其AI代理。

Although the internal details of the AI system each team within the AlphaDogFight are not made public, we can deduce the techniques that are utilized by referring to research within the areas of using machine learning and neural networks for air combat and simulations.

尽管AlphaDogFight中每个团队的AI系统内部细节尚未公开，但我们可以通过参考在使用机器学习和神经网络进行空战和模拟领域内的研究来推断所使用的技术。

AI的以下小节涵盖了一些利用的AI技术： (Some AI techniques utilized are encompassed in the following subsections of AI:)

Deep Learning
深度学习
Reinforcement Learning
强化学习
Deep Neural Networks
深度神经网络
Artificial Neural Networks
人工神经网络
Heuristics based algorithms
基于启发式的算法

空战系统AI的领域可以应用于： (Areas of air combat systems AI can be applied to:)

Path Planning
路径规划
Navigation
导航
Decision making
做决定
Evasion
逃避
Collision detection
碰撞检测
Tracking and following
追踪与追踪

Since the early 90s, we’ve seen research that explored the utilization of neural networks for the task of decision making.

自90年代初以来，我们已经看到了研究探索将神经网络用于决策任务的研究。

In the case of dogfights, there have been efforts that have explored the use of neural networks for fighter jets path prediction in aerial based combat.

在混战的情况下，已经进行了一些努力来探索神经网络在空中战斗中对战斗机路径预测的使用。

In the previous decade, neural networks have been shown to have the capability to learn manoeuvres when equipped with a set of heuristic-based knowledge and expertise.

在过去的十年中，神经网络在配备了一系列基于启发式的知识和专业知识后，具有学习操纵的能力。

More recently, deep learning techniques have provided the capability to solve computer vision task such as object detection, object recognition and tracking in a real-time environment, with high accuracy.

最近，深度学习技术已经提供了以高精度解决计算机视觉任务(例如在实时环境中进行对象检测，对象识别和跟踪)的能力。

AI agent within dogfights can be equipped with deep learning techniques that enable the detection and monitoring of an opponent agent.

缠斗中的AI代理可以配备深度学习技术，以检测和监视对手代理。

Perception within AI agents can be achieved through the utilization of deep learning techniques that can identify and track objects within a stream of pixels from the simulated environment.

可以通过利用深度学习技术来实现AI代理内部的感知，深度学习技术可以识别和跟踪来自模拟环境的像素流中的对象。

Some AI agent can even be equipped with obstacle avoiding procedures if the environment is perceived through methods such as raycasting.

如果通过光线投射等方法感知环境，则某些AI代理甚至可以配备避障程序。

Teams within the DARPA’s AlphaDogfight have built an end to end system for their solution; this means that they have developed not only an AI agent that can fly an F-16, but they’ve also developed the simulation environment the AI-agent is trained and developed, amongst other components.

DARPA的AlphaDogfight团队已经为他们的解决方案构建了一个端到端系统。这意味着他们不仅开发了可以飞行F-16的AI代理，而且还开发了AI代理经过训练和开发的仿真环境，以及其他组件。

There was a research conducted in 2014 by academics based in the Netherlands. This research explored the utilization of a machine learning technique known as dynamic Scripting for the development of opponent agent behavioural systems within air combat simulation.

2014年，荷兰的学者进行了一项研究。这项研究探索了一种称为动态脚本的机器学习技术在空战模拟中用于开发敌方特工行为系统的方法。

Dynamic Scripting(DS) is typically found in gaming environments. DS is derived from reinforcement learning, where agents adaptively navigate through a set of predefined behaviours and rules until a goal is achieved.

动态脚本(DS)通常在游戏环境中找到。 DS是从强化学习中获得的，在强化学习中，代理通过一组预定义的行为和规则自适应地导航，直到实现目标为止。

Within an air combat simulated environment, you could have enemy fighter jets (Computer Generated Forces) that have the goal of evading or shooting down other agents. These enemy fighter jets will traverse through a set of actions and rules that are have assigned weights which enables the AI agents to learn the optimal behaviours within a scenario.

在空战模拟环境中，您可能拥有旨在逃避或击落其他特工的敌方战斗机( 计算机生成的部队 )。这些敌方战斗机将穿越一组分配有权重的动作和规则，使AI代理能够了解场景中的最佳行为。

Neural networks have proven to be very useful in the development of AI-based systems that can handle the navigation and control systems of aircraft.

神经网络已被证明在基于AI的系统开发中非常有用，该系统可以处理飞机的导航和控制系统 。

A large portion of the capabilities of AI systems developed for the AlphaDogFight will be around manoeuvres. Artificial Neural Networks have been applied to the creation of AI agents that have adaptive manoeuvring capabilities through the utilization of self-organizing neural networks (Self Organizing feature maps).

为AlphaDogFight开发的AI系统的大部分功能将围绕机动性进行。人工神经网络已被用于通过利用自组织神经网络来创建具有自适应操纵能力的AI代理 (自组织功能图)。

Research conducted by academics Xianbing Zhang et al. showcased the use of Q-networks (derived from Q-learning reinforcement learning algorithm) to optimize the search for manoeuvre decisions within air combat systems.

研究的学者张陷饼等进行。展示了使用Q网络 (源自Q学习强化学习算法)来优化空战系统内机动决策的搜索。

One of the critical aims of AlphaDogFight is to increase trust in AI-based systems in air combat.

AlphaDogFight的关键目标之一是增强对空战中基于AI的系统的信任。

Shifting the responsibilities of decision making in-flight scenarios from the human pilot to an automated intelligent system is one method of developing trust in AI.

将飞行中决策场景的职责从人工飞行员转移到自动化智能系统是一种发展对AI信任的方法。

A pilot has to make high cognitive decisions when flying F-16 or other jet fighters, and there are some notable benefits in reducing or completely relieving the pilot of decision-making responsibilities.

飞行员在驾驶F-16或其他喷气式战斗机时必须做出较高的认知决策，并且在减少或完全减轻飞行员的决策责任方面有一些明显的好处。

结论 (Conclusion)

There are significant efforts within the academic and industrial sectors that are exploring the full applicability of AI systems within autonomous flight systems and air combat systems.

学术界和工业界在探索AI系统在自动飞行系统和空战系统中的完全适用性方面做出了巨大的努力。

Regardless of the outcome of AlphaDogFight, you should be aware of the amount of effort and resources that are directed at developing intelligent AI systems for combat.

不管AlphaDogFight的结果如何，您都应该意识到用于开发用于战斗的智能AI系统的精力和资源。

There are already some proven methods that depict a future where AI systems can at least handle a significant portion of the roles and decisions that are involved in flight control and air combat systems.

已经有一些行之有效的方法描绘了一个未来，人工智能系统至少可以处理飞行控制和空战系统中涉及的大部分角色和决策。