11-19 更新 实时
首页/ 深度/ 正文

MIT 微型飞行机器人借 AI 提速 450%

深度 来源:ScienceDaily AI 2026-09-22 20:52 0 阅读 机器翻译
MIT 微型飞行机器人借 AI 提速 450%

来自研究机构

MIT 的小型飞行机器人借助 AI 速度提升 450%

MIT 的小型 AI 飞行机器人如今几乎能像昆虫一样运动,可在短短 11 秒内完成快速转向和 10 次翻转。

日期:2026 年 9 月 22 日来源:麻省理工学院摘要:一套新的 AI 控制系统让 MIT 的小型飞行机器人具备了昆虫般的敏捷性,速度提升约 450%,并能在 11 秒内完成 10 次空翻。这项技术最终有望让微型机器人搜索地震废墟,并进入常规无人机无法到达的危险空间。分享:

左图:一张延时照片展示了一个飞行微型机器人正在做翻转动作。右图:Chen 的团队研发机器人昆虫已超过五年。图片来源:MIT 软体与微型机器人实验室

有朝一日,小型飞行机器人或许能帮助救援人员搜寻地震后被困在倒塌建筑物下的人员。由于体型小巧,这些机器人昆虫有可能穿过大型无人机无法进入的狭窄空间,同时避开墙壁、瓦砾和坠落物体。

然而直到最近,空中微型机器人的敏捷性仍远不及启发它们设计的昆虫。它们通常移动缓慢,飞行路径也相对简单。

MIT 的研究人员如今展示了一种新方法,让昆虫尺度的飞行机器人获得了大幅提升的速度与敏捷性。该团队开发了一套基于 AI 的控制器,使这款机器人昆虫能够完成高难度的空中机动动作,包括连续的身体翻转。

研究人员采用了一套由两部分组成的控制系统,旨在兼顾性能与计算效率,与先前的最佳结果相比,将机器人的速度提升了约450%,加速度提升了约250%。

该机器人足够敏捷,能在11秒内完成10个连续空翻,即便有风扰试图将其吹离预定路径。

“我们希望能在那些传统四旋翼机器人难以飞入、而昆虫却能够穿行的场景中使用这类机器人。如今,借助我们受生物启发的控制框架,我们的机器人在速度、加速度和俯仰角方面的飞行性能已可与昆虫相媲美。这是朝着这一未来目标迈出的相当令人振奋的一步。”麻省理工学院电气工程与计算机科学系(EECS)副教授、电子学研究实验室(RLE)软体与微型机器人实验室负责人、该机器人论文的共同资深作者 Kevin Chen 说道。

该论文的共同第一作者还包括:MIT EECS 研究生 Yi-Hsuan Hsiao;Andrea Tagliabue(24 届博士);航空航天系(AeroAstro)研究生 Owen Matteson;以及 EECS 研究生 Suhan Kim;Tong Zhao(23 届工程硕士);以及共同资深作者 Jonathan P. How,他是航空航天系 Ford 工程学教授、信息与决策系统实验室(LIDS)的首席研究员。该研究发表在《Science Advances》上。

AI 让机器人拥有更智能的飞行控制器

Chen 的团队已用五年多时间研发机器昆虫。

该团队最近制造出他们微型机器人的更耐用版本,其大小约相当于一枚微型盒式磁带,重量不到一枚回形针。这个更新的设计拥有更大的扑翼,能够支持更敏捷的飞行。翅膀由柔性人工肌肉驱动,其收缩速度足够快,可产生极快的振翅。

物理设计有所改进,但机器人的控制器仍是主要限制。该控制器充当机器人的“大脑”,判断机器人所处位置并决定其应如何移动。在更早的版本中,必须由人工手动调校控制器。

要让机器人以真正昆虫般的速度和激进程度飞行,研究人员需要一个能够处理不确定性、同时快速求解复杂控制问题的系统。

一个强大到足以做到这一点的控制器,通常需要过多计算量,难以实时运行,尤其是因为如此轻量的飞行器的空气动力学极为复杂。

为解决这一问题,Chen 的团队与 How 的团队合作,开发了一套两步式 AI 驱动的控制系统。该设计兼具应对高难度高速机动所需的鲁棒性,以及足以实时运行的计算效率。

“硬件的进步推动了控制器的发展,使我们在软件层面能做的事情更多;但与此同时,随着控制器的发展,他们在硬件上也能做更多事情。当 Kevin 的团队展示出新能力时,我们就展示出我们能够利用这些能力,”How 说。

教微型机器人规划高难度机动动作

该系统第一部分使用所谓模型预测控制器。这种控制器依赖动态数学模型来预测机器人将如何行动,并确定安全沿目标飞行路径飞行的最佳动作序列。

尽管这种方法需要大量算力,但它可以规划空中翻转、急转弯以及机体角度剧烈变化等困难动作。该规划器还会考虑机器人能产生的力和力矩的上限,帮助其避免可能导致坠毁的动作。

连续翻转尤其具有挑战性。要完成一个接一个的空翻,机器人必须以完全正确的方式减速,从而在合适的条件下开始每一次新的翻转。

“如果混入微小误差,而你还试图带着这些微小误差把那个翻转重复 10 次,机器人就会直接坠毁。我们需要稳健的飞行控制,”How 说。

随后,研究人员用这位专家规划器,通过名为模仿学习的过程,训练出一个基于深度学习模型构建的“策略”。该策略充当机器人的实时决策系统,决定它应如何飞、飞向哪里。

实际上,模仿学习让团队得以将计算需求更高的控制器的行为,捕捉到一个更快的 AI 模型中,其运行速度足以支持实时飞行。

挑战的一大重点在于生成足够多的有用训练数据,以教会该策略如何处理高难度机动,同时又不至于创建不必要的大规模数据集。

“这套稳健的训练方法是这项技术的秘诀所在,”How 解释道。

飞行过程中,基于 AI 的策略接收有关机器人位置的信息,并将其转换为实时指令,用于控制推力、扭矩等因素。

昆虫般的速度与敏捷性

实验表明,这套两步控制系统大幅提升了机器人的性能。这台昆虫尺度的机器飞行速度提高了 447%,加速度提升了 255%。

它还在仅 11 秒内完成了 10 个空翻,同时与预定飞行路径的偏差保持在约 4 至 5 厘米以内。

“这项工作表明,传统上速度受限的软体机器人和微型机器人,如今可以借助先进的控制算法,实现接近自然昆虫和更大机器人的敏捷性,为多模态运动开辟了新的机会,”Hsiao 表示。

团队还演示了一种被称为扫视(saccade)的动作。在昆虫身上,这一动作包括急剧俯仰身体、迅速移动到新位置,然后向相反方向俯仰以制动。这种快速加速与制动有助于昆虫判断自身位置并保持清晰视觉。

“这种仿生飞行行为在未来我们开始为机器人搭载摄像头和传感器时,可能会对我们有所帮助,”Chen 表示。

迈向自主飞行微型机器人

未来研究的一个主要目标是为这些微型机器人配备机载摄像头和传感器。这最终可能使它们能够在户外飞行,而不必依赖复杂的外部动作捕捉系统。

该团队还计划研究机载传感是否能够帮助成群的这类机器人在导航时避免相互碰撞并协调各自的运动。

“对于微型机器人领域,我希望这篇论文标志着一个范式转变,因为它表明我们可以开发出一种同时具备高性能和高效率的新型控制架构,”Chen 表示。

这项研究的部分资金来自美国国家科学基金会(NSF)、海军研究办公室、空军科学研究办公室、MathWorks 以及 Zakhartchenko Fellowship。

材料由麻省理工学院提供。注:内容可能因风格和篇幅而经过编辑。

Yi-Hsuan Hsiao, Andrea Tagliabue, Owen Matteson, Suhan Kim, Tong Zhao, Jonathan P. How, YuFeng Chen. Aerobatic maneuvers in insect-scale flapping-wing aerial robots via deep-learned robust tube model predictive control. Science Advances, 2025; 11 (49) DOI: 10.1126/sciadv.aea8716

原文较长,译文为节选;译文由英文原文机器翻译,术语按行业惯例统一,可在文末展开原文核对。

查看英文原文(MIT’s tiny flying robot gets 450% faster with AI)

MIT’s tiny flying robot gets 450% faster with AI | ScienceDaily

from research organizations

MIT’s tiny flying robot gets 450% faster with AI

MIT’s tiny AI-powered flying robot can now move almost like an insect, performing rapid turns and 10 flips in just 11 seconds.

Date:September 22, 2026Source:Massachusetts Institute of TechnologySummary:A new AI control system lets MIT’s tiny flying robot move with insect-like agility, boosting its speed by about 450 percent and allowing it to pull off 10 somersaults in 11 seconds. The technology could eventually enable miniature robots to search earthquake rubble and navigate dangerous spaces that conventional drones cannot reach.Share:

MIT’s Tiny Flying Robot Gets 450% FasterLeft: A time-lapse photo shows a flying microrobot performing a flip. Right: Chen’s group has been building robotic insects for more than five years. Credit: MIT Soft and Micro Robotics Laboratory

Tiny flying robots may one day help rescuers search for people trapped beneath collapsed buildings after earthquakes. Because of their small size, these robotic insects could potentially move through narrow spaces that larger drones cannot enter while avoiding walls, debris, and falling objects.

Until recently, however, aerial microrobots have been far less nimble than the insects that inspired them. They typically moved slowly and followed relatively simple flight paths.

Researchers at MIT have now demonstrated a new approach that gives an insect-scale flying robot much greater speed and agility. The team developed an AI-based controller that allows the robotic bug to perform demanding aerial maneuvers, including repeated body flips.

Using a two-part control system designed to balance performance with computational efficiency, the researchers increased the robot's speed by about 450 percent and its acceleration by about 250 percent compared with their previous best results.

The robot was agile enough to complete 10 consecutive somersaults in 11 seconds, even while wind disturbances tried to knock it away from its intended path.

"We want to be able to use these robots in scenarios that more traditional quadcopter robots would have trouble flying into, but that insects could navigate. Now, with our bioinspired control framework, the flight performance of our robot is comparable to insects in terms of speed, acceleration, and the pitching angle. This is quite an exciting step toward that future goal," says Kevin Chen, an associate professor in the Department of Electrical Engineering and Computer Science (EECS), head of the Soft and Micro Robotics Laboratory within the Research Laboratory of Electronics (RLE), and co-senior author of a paper on the robot.

Chen is joined on the paper by co-lead authors Yi-Hsuan Hsiao, an EECS MIT graduate student; Andrea Tagliabue PhD '24; and Owen Matteson, a graduate student in the Department of Aeronautics and Astronautics (AeroAstro); as well as EECS graduate student Suhan Kim; Tong Zhao MEng '23; and co-senior author Jonathan P. How, the Ford Professor of Engineering in the Department of Aeronautics and Astronautics and a principal investigator in the Laboratory for Information and Decision Systems (LIDS). The research was published in Science Advances.

AI Gives the Robot a Smarter Flight Controller

Chen's group has spent more than five years developing robotic insects.

The team recently created a more durable version of their tiny robot, which is about the size of a microcassette and weighs less than a paperclip. This newer design has larger flapping wings that support more agile flight. The wings are driven by soft artificial muscles that contract rapidly enough to produce extremely fast wingbeats.

The physical design had improved, but the robot's controller remained a major limitation. This controller acts as the robot's "brain," determining where the robot is and deciding how it should move. In earlier versions, a human had to tune the controller by hand.

For the robot to fly with the speed and aggressiveness of a real insect, the researchers needed a system that could handle uncertainty while rapidly solving complex control problems.

A controller powerful enough to do that would normally demand too much computation to operate in real time, especially because the aerodynamics of such a lightweight flying machine are highly complex.

To solve this problem, Chen's group collaborated with How's team to develop a two-step AI-driven control system. The design combines the robustness needed for difficult, high-speed maneuvers with enough computational efficiency to work in real time.

"The hardware advances pushed the controller so there was more we could do on the software side, but at the same time, as the controller developed, there was more they could do with the hardware. As Kevin's team demonstrates new capabilities, we demonstrate that we can utilize them," How says.

Teaching a Tiny Robot to Plan Difficult Maneuvers

The first part of the system uses what is known as a model-predictive controller. This type of controller relies on a dynamic mathematical model to predict how the robot will behave and determine the best sequence of actions for safely following a desired flight path.

Although this method requires substantial computing power, it can plan difficult movements such as aerial flips, sharp turns, and aggressive changes in body angle. The planner also takes into account limits on the amount of force and torque the robot can produce, helping it avoid movements that could lead to a crash.

Repeated flips are particularly challenging. To perform one somersault after another, the robot must slow down in exactly the right way so that it begins each new flip under the proper conditions.

"If small errors creep in, and you try to repeat that flip 10 times with those small errors, the robot will just crash. We need to have robust flight control," How says.

The researchers then used this expert planner to train a "policy" built on a deep-learning model through a process called imitation learning. The policy serves as the robot's real-time decision-making system, determining how and where it should fly.

In effect, imitation learning allows the team to capture the behavior of the more computationally demanding controller in a faster AI model that can operate quickly enough for real-time flight.

A major part of the challenge was generating enough useful training data to teach the policy how to handle aggressive maneuvers without creating an unnecessarily large data set.

"The robust training method is the secret sauce of this technique," How explains.

During flight, the AI-based policy receives information about the robot's position and converts it into real-time commands controlling factors such as thrust force and torques.

Insect-Like Speed and Agility

Experiments showed that the two-step control system dramatically improved the robot's performance. The insect-scale machine flew 447 percent faster and achieved a 255 percent increase in acceleration.

It also completed 10 somersaults in only 11 seconds while remaining within about 4 or 5 centimeters of its planned flight path.

"This work demonstrates that soft and microrobots, traditionally limited in speed, can now leverage advanced control algorithms to achieve agility approaching that of natural insects and larger robots, opening up new opportunities for multimodal locomotion," says Hsiao.

The team also demonstrated a movement known as a saccade. In insects, this involves pitching the body sharply, rapidly moving to a new position, and then pitching in the opposite direction to stop. This quick acceleration and braking can help insects determine their position and maintain clear vision.

"This bio-mimicking flight behavior could help us in the future when we start putting cameras and sensors on board the robot," Chen says.

Toward Autonomous Flying Microrobots

One major goal for future research is to equip the microrobots with onboard cameras and sensors. That could eventually allow them to fly outdoors without depending on a sophisticated external motion capture system.

The team also plans to investigate whether onboard sensing could help groups of the robots avoid crashing into one another and coordinate their movements while navigating.

"For the micro-robotics community, I hope this paper signals a paradigm shift by showing that we can develop a new control architecture that is high-performing and efficient at the same time," says Chen.

This research is funded, in part, by the National Science Foundation (NSF), the Office of Naval Research, Air Force Office of Scientific Research, MathWorks, and the Zakhartchenko Fellowship.

Materials provided by Massachusetts Institute of Technology. Note: Content may be edited for style and length.

Yi-Hsuan Hsiao, Andrea Tagliabue, Owen Matteson, Suhan Kim, Tong Zhao, Jonathan P. How, YuFeng Chen. Aerobatic maneuvers in insect-scale flapping-wing aerial robots via deep-learned robust tube model predictive control. Science Advances, 2025; 11 (49) DOI: 10.1126/sciadv.aea8716

Massachusetts Institute of Technology. "MIT’s tiny flying robot gets 450% faster with AI." ScienceDaily. ScienceDaily, 22 September 2026. .

Massachusetts Institute of Technology. (2026, September 22). MIT’s tiny flying robot gets 450% faster with AI. ScienceDaily. Retrieved November 5, 2026 from www.sciencedaily.com/releases/2026/09/260921081114.htm

Massachusetts Institute of Technology. "MIT’s tiny flying robot gets 450% faster with AI." ScienceDaily. www.sciencedaily.com/releases/2026/09/260921081114.htm (accessed November 5, 2026).

原文为英文,正文由自动抓取并保留原文。

本文由采集器自动同步

原文发布于 ScienceDaily AI(MIT’s tiny flying robot gets 450% faster with AI),标题、摘要与正文由机器翻译,术语按行业惯例统一,未经人工逐句校订。

阅读原文 ↗