RoboSpeak — WeChat (mp_account, tier=T2)

“

Solving the long-standing problem of flight robots losing stability with a single collision.

Quadcopter flight robots can fly and take photos, but have long been hesitant to engage in direct physical interaction with the environment.

Aerial photography, inspection, and cargo delivery are already well-established basic tasks for them. In these scenarios, they only need to hover, move, and take photos, maintaining a distance from the surroundings throughout. The real challenge lies in actively engaging in physical interaction with the environment, such as catching a moving target, hanging on a wall, shattering ice layers, or retrieving objects from water.

Although these tasks appear very different, the root cause of their difficulties is the same. A quadcopter is an underactuated system that maintains balance through continuous fine adjustments. When it exerts force, the reaction force will violently tilt the body along the mechanical arm.

In other words,

For flight robots, exerting force and maintaining stable hovering are inherently contradictory. This is the inescapable dilemma in the field of aerial interaction.

The team led by He Bin and Jiang Yongkang at the Shanghai Autonomous Intelligent Unmanned System Science Center at Tongji University adopted a new approach to solving the problem. They do not let the flight control system directly withstand sudden impacts, instead breaking the entire action into two independent tasks.

The asymmetric bistable pulse module is responsible for “quick strike”, while the serial flexible arm handles “impact digestion”, performing attenuation before the disturbance reaches the fuselage.

Related achievements recently appeared on the authoritative robotics journal Science Advances.

This research's biomimicry mimics the hunting strategy of raptors—keeping the body steady while the limbs instantly extend forward to capture prey. The platform itself remains a quadcopter drone.

The biomimetic flying robot is equipped with pulsed limbs, enabling dynamic aerial interaction with morphological buffering.

01.

Energy storage launch + flexible energy dissipation, resolving the dilemma of aerial interaction

Existing aerial manipulators struggle to balance burst power and stability. Rigid arms offer good positioning accuracy, but the impact from high-speed contact is directly transmitted to the body, easily disrupting the hover state; flexible arms have inherent shock absorption, but their response speed is insufficient for dynamic grasping or ice-breaking tasks. Many fast-deploying mechanisms, although quick to extend, rely on external drives such as pneumatic or magnetic forces, making it difficult to adapt to lightweight onboard platforms.

Flight control seeks stability, while interaction actions require instantaneous bursts—two demands that are hard to coexist in a single mechanism. The team's solution is functional decoupling, allowing different structures to perform their respective roles.

The pulsed module is dedicated to energy storage and completing the launch, while the connected flexible arm is responsible for dissipating impact. Flight control no longer has to resist sudden disturbances, focusing instead on positioning, alignment, and trigger judgment.

The core of the pulsed module is a piece of polyimide flexible beam with a thickness of 0.7–0.9 mm, which remains in a metastable state to accumulate elastic potential energy.

The motor slowly pushes the slider to continuously store energy. Once it crosses the critical threshold, the structure instantly jumps to release energy, propelling the entire arm to rapidly extend.

The non-symmetric bistable configuration is the key to this design. Traditional symmetric bistable structures require higher energy storage, leading to greater trigger force, which is difficult for onboard small motors to drive; the asymmetric configuration decouples energy storage capacity from the trigger threshold, and with geometric pre-loading, the structure's initial state is close to the critical point, reducing the trigger resistance.

The trade-off is equally clear. When the beam thickness increases from 0.7 mm to 0.9 mm, the peak trigger force rises from 20 N to over 45 N, with energy release increasing by about 2.5 times. Engineering design requires finding a balance point between energy storage demand and motor load.

The flexible arm is composed of a rigid ring frame and a bistable PET thin film with a thickness of 0.15–0.25 mm.

Rigid supports transfer the kinetic energy of the projectile, achieving a domino-style step-by-step deployment; PET thin film undergoes deformation and dissipates impact upon contact with the target, and the paper defines this as the 'mechanical impact filter'.

The end effector eliminates the need for independent drive commands, relying on contact force to directly lock. The arm body reset relies on SMA shape memory alloy springs, rather than motors, and can complete contraction and reset under 7.4 V voltage and 0.5 s pulse heating.

Mechanical structure and geometric design of the pulse module.

02.

70 milliseconds strike, achieving 76.46% impact attenuation

Under test configuration, the arm retracts from 71 mm to 166 mm, an extension ratio of approximately 2.3 times, with the entire process taking only 70 ms, and the end-effector peak velocity of 2.1 m/s; reset requires 440 ms, with striking much faster than retraction.

Kinematic characteristics of the pulse-driven cycle. (A) Ballistic extension phase. Under the trigger of the pulse module, the tested flexible arm fully extends within 70 milliseconds, with a peak velocity of 2.1 meters per second. (B) Thermal recovery phase driven by shape memory alloy (SMA). The embedded SMA actuator returns the flexible arm to a compact state within 440 milliseconds.

Impact data is the core highlight of this work. Under the same contact conditions, 3-section flexible arm compared to rigid arm: peak reaction force drops from 13.92 N to 9.22 N, 33.76%; the impulse in the first 100 ms drops from 411.20 to 96.80 mN·s, a decrease of 76.46%; the peak body displacement also drops by 74.57%.

The 76.46% corresponds to the impulse integral in the first 100 milliseconds, i.e., the total impact, rather than the instantaneous peak impact force. A rigid arm would transmit all the impact directly to the body, while a flexible arm dissipates most of the disturbance in advance, reducing the total impact load by nearly three-quarters.

Another set of experiments further demonstrates the value of morphological intelligence.

A 7-section flexible arm, during fixed platform testing, achieves a passive adaptation range of 277 mm, which is 70% of the total length of 395 mm. After launch, the control system no longer calculates the arm length in real time, and the arm body continuously extends forward until it contacts the target, relying on the structure to achieve distance self-adaptation.

Autonomous target interception with passive arm adaptability.

03.

Four types of real flight scenarios have been successfully tested: able to grasp, stop efficiently, break through, and retrieve.

The research team conducted four types of real flight scenarios, with a wide range of applications.

Dynamic grasping

using a 3-section arm, in

three vertical distances of 16.8 cm, 19.2 cm, and 24.0 cm, all completed effective contact.

Onboard YOLOv5n identifies the target, depth camera outputs pose; when horizontal error is less than 1 cm, launch is initiated, and the remaining vertical error is handled by the arm's passive tolerance, with the test target being a snake-like moving model.

Fixed-point stop

using an upward deployment configuration, relying on geometric tolerance to adapt to height, without the need for high-precision position closed-loop locking; after fixation, the drive motor can shut down to save power, suitable for long-term fixed-point monitoring.

Ice-breaking scenario

is the domain of the pulse module, relying on instantaneous ballistic impact to break through ice layers, while the flexible arm synchronously absorbs the transient impact load, reducing the load transmitted to the body.

Cross-medium grasping

is divided into two operating conditions. Air-to-water grasping, the arm body probes into the water surface, with the body hovering above the water; gel medium grasping, contact time is controlled within 0.2 seconds, relying on SMA rapid retraction to reduce the risk of adhesive drag in viscous media.

Pulse drive and morphological buffering enable multifunctional aerial physical interaction. (A) Using the pulse module to break ice. The instantaneous impact force can shatter the ice layer, while the flexible arm acts as a mechanical impact filter, reducing the collision load transmitted to the flight platform. (B) Cross-medium recovery (air-to-water grasping). Utilizing a larger extension ratio, the drone can recover the target from a greater distance. A larger separation distance allows the drone to stay away from the water interaction region, thus supporting cross-medium recovery. (C) Rapid viscous medium sampling (air-to-gel grasping). A contact time of less than 0.2 seconds supports a rapid submersion-retraction cycle for recovering the target from high-damping media.

The paper also honestly acknowledges limitations. During experiments, it is still observable that rotor downwash airflow causes interference, and the current solution can only reduce the impact by increasing the distance, but cannot completely eliminate it.

04.

Delegating some control to the structure

Traditional high-speed aerial interaction requires continuous closed-loop control of height to ensure precise contact with the target. This design changes the approach: after launch, it abandons real-time control of arm length, and the final working length is defined by the contact event.

For flight control, it reduces one control variable that requires high-frequency adjustment; for the mechanical body, it inherently provides a layer of passive tolerance. The paper's summary is very concise: morphological buffering, Shift the bottleneck of aerial physical interaction from algorithm control to embodied mechanical design.

Modularity is another aspect of this design. A modular architecture supports adding or removing arm segments to adjust the working range, but the number of segments is a double-edged sword: longer stroke will simultaneously increase the demand for launch energy, structural self-weight, and disturbance torque. The thickness of the thin sheet is increased from 0.15 mm to 0.25 mm, raising the energy barrier from 10 mJ to 65 mJ, increasing structural stiffness, and also raising the trigger threshold. All these parameters require comprehensive trade-offs.

This pulse arm does not transform the quadcopter into a bird. It draws on the division logic of how a raptor hunts, keeping the body steady while the limbs strike instantly, with collision disturbances being absorbed by the limbs themselves.

The contradiction between speed and stability, a long-standing issue, does not necessarily have to be solved by iterating more complex control algorithms. Sometimes, a single flexible rod that can store energy and deform can alleviate the pressure on the control system.

Looking at the entire mobile robot industry, this is also a signal worth noting. The industry has long relied on stronger computing power and denser closed-loop systems to counteract environmental disturbances, while morphological intelligence offers another approach—decentralizing some perception, fault tolerance, and dynamics regulation capabilities into the mechanism itself.

For platforms such as aerial operations and mobile grasping, which are highly constrained in terms of computational power, payload, and endurance, having the structure inherently “understand fault tolerance” first is often more practical than endlessly stacking controller performance.

Paper link: https://www.science.org/doi/10.1126/sciadv.aeg4557

END

Industrial robot companies

Eston Automation

|

Efte Robot

|

Fao Robot

|

Yuexiang Robot

|

Jieka Robot

|

Songling Robot

|

Luoshi Robotics

|

A Tongmu Robot

|

JiZhiJia

|

Hikvision Robot

|

Yifei Technology

|

Ailite Robot

Service and special robot companies

Yijiahe

|

Jingpin Tieshang

|

Qitem Robotics

|

Shihe Robot

|

Pudu Robot

|

Shiluo Robot

|

Kuma Technology MAMMOTION

Humanoid robot companies

UBTech Robotics

|

Unitree

|

Yunchedu

|

Xingdongjinyuan

|

Weijing Robot

|

Zhujidongli

|

Leju Robot

|

Daxiang Robot

|

Magic Atom

|

Zhongqing Robot

|

Pasi Ni Perception

|

Cyborg Robot

|

Digital Huaxia

|

Fourier Intelligent

|

Tianlian Robot

|

Kepler Humanoid Robot

|

Lingbao CASBOT

|

Qingbao Robot

|

Zhejiang Humanoid Robot Innovation Center

|

Dongyi Technology

|

Zhishi Technology

|

PNDbotics

|

Zhuoyide Robot

|

Qinglang Intelligent

|

Galileo GALILEO

|

Songyan Power

|

Tianji Intelligent

|

Woaan Robot

|

LiGong Huahui

|

Jiace Jinhua

Embodied intelligence companies

Cross Dimension Intelligent

|

Galaxy Universal

|

Qianxun Intelligent

|

Lingxin Qiaoshou

|

Ruierman Intelligent

|

Microintelligence

|

Tuisheng Technology

|

Zhongke Silicium Records

|

Shutu Technology

|

Lingqiao Intelligent

|

Xingchen Intelligent

|

Qiongche Intelligent

|

Fangzhou Infinite

| iFLYTEK |

Beijing Humanoid Robot Innovation Center

|

National and Local Co-built Humanoid Robot Innovation Center

|

Daimeng Robot

|

Shibitu Robot

|

Xinghaitu

|

Yuequan Bionics

|

Zeroth Power Robot

|

Zhongke ShenGu

|

Zhi Fangping

|

Daka Robot

|

Haocun Technology

|

Ju Shi Intelligent

|

Xynova Xiner Future

|

Feixi Technology

|

Weilu Power

|

Boden Intelligent

|

Qianjue Technology

|

Lingsheng Technology

|

Ji Cui Intelligent Manufacturing

|

Xinbeit Technology

|

Chen Hunxian Technology

|

Dexmal Original Power

|

Youliqi

|

Self-variable

|

Ruiyan Zhikong Lingqiao Hand

|

Qiwu Technology

|

RoboScience Machine Science

|

Zhongke Fifth Dynasty

|

Critical Point

|

Danghong Technology

|

Qiaojie Shu Wu

|

Vbot Veta Power

|

Tashan Technology

|

Ju Brain Pan Shi

|

Youai Zhijhe Robot

|

Zhixing Jian

|

Aimiao Robot

|

Lu Ming Robot

|

Quantum Intelligent

|

Ruoyu Technology

|

Tiehu Robot

|

Zhi Wuji

|

Origin Infinite

Medical robot companies

Yuanhua Intelligent

|

Tianzhihang

|

Sizhe Rui Intelligent Medical

|

Jingfeng Medical

|

Tuo Dao Medical

|

Zhenyida

|

Shuru® Robot

|

Luosenbot

|

Shuimu Dongfang

｜

Kangnuo Sistem

|

Disight Medical

Upstream industry chain companies

Lvde Harmonic

|

Yinshi Robotics

|

Kunwei Technology

|

Muta Intelligent

|

Qingpou Vision

|

Benmo Technology

|

Lan Dian Touch Control

| Xinjingcheng Sensor

|

BrainCo Strong Brain Technology

|

Yuli Instrument

|

Jiya Jingji

|

Sila Technology

|

Shenyuan Sheng

|

Feipu Navigation Technology

|

Inks

|

Juxi Intelligent Drive

|

Lingyun Guang Yuanke Vision

|

Xuanji Power

|

Yiyou Technology

| Ruiyuan Precision |

Lingzhu Era

|

HIT Huawei Ke

|

Xinghui Sensor

|

Lingdi Technology

|

Quan Zhi Bo

|

CubeMars Robot Power

|

Wanglong Robot Elevator

|

Hangkai Microelectronics

|

Wutong Sensing and Control