A 26-Gram Butterfly-Inspired Robot Achieving Autonomous Tailless Flight

This paper introduces \textit{AirPulse}, a 26-gram butterfly-inspired robot that achieves the first autonomous, closed-loop tailless flight at this scale by replicating low-frequency, high-amplitude biomechanical traits through a hierarchical control architecture featuring Stroke Timing Asymmetry Rhythm (STAR).

The Gist

Imagine trying to build a robot that flies like a butterfly. Most flying robots today are like helicopters or birds: they have rigid wings, spin or flap very fast, and rely on a tail or extra fins to stay steady. But butterflies are different. They are like clumsy dancers who wobble, undulate, and flap their wings slowly, yet somehow manage to zip around, climb steeply, and turn on a dime without ever falling.

For a long time, engineers thought this "wobbly" style of flight was too chaotic to control with a computer. That is, until a team from Tsinghua University built AirPulse.

Here is the story of AirPulse, explained simply.

1. The Tiny, Wobbly Wonder

AirPulse is a tiny robot, weighing just 26 grams (about the weight of a large chocolate bar or a few coins). It is the first robot of its size to fly on its own, with all its "brain" and sensors built right inside its body, without any wires tethering it to the ground.

Unlike other flapping robots that try to be stiff and precise, AirPulse embraces the butterfly's natural chaos. It has:

• Low-aspect-ratio wings: Think of them as short and wide, like a butterfly's, rather than long and thin like a hawk's.
• Compliant (flexible) wings: Instead of being made of hard plastic, its wings are like delicate fabric reinforced with tiny carbon-fiber "veins." This lets them bend and twist in the wind, just like a real butterfly.
• No tail: It has no tail to steer with. It has to figure out how to turn using only its two wings.

2. The "Wobble" Problem

When a real butterfly flaps its wings, its body doesn't just stay still; it bobs up and down and twists side-to-side. This is called body undulation.

In a normal airplane, the pilot wants the plane to be smooth. In a butterfly, the "wobble" is actually part of the flight. As the wings flap, they shift the robot's center of gravity and change how heavy it feels to spin.

• The Analogy: Imagine trying to ride a unicycle while someone is constantly pushing you from the side and changing the weight of the seat every second. Most robots would crash. AirPulse, however, learns to dance with the wobble rather than fighting against it.

3. The Secret Sauce: The "STAR" Rhythm

To control this chaotic dance, the engineers invented a new mathematical trick called STAR (Stroke Timing Asymmetry Rhythm).

Think of flapping your wings like a metronome.

• Old way: To turn left, you might try to flap your left wing harder or change the angle of your wings. This is like trying to steer a car by hitting the brakes on one side. It's jerky and unstable.
• The STAR way: Instead of changing how hard you flap, you change when you flap. You make the left wing's "down" stroke happen a tiny bit faster than the "up" stroke, while the right wing does the opposite.

The Metaphor: Imagine two people walking side-by-side. If they both walk at the same speed, they go straight. If one person takes a slightly longer step forward and a shorter step back, they start to curve. STAR is the robot's way of telling its wings to take "longer steps" at the perfect moment to steer smoothly, without jerking the robot around.

4. The Brain: Dancing with the Wind

The robot's computer (its brain) has to work very fast. It can't just look at where the robot is now; it has to predict where it will be in the next split second because the robot is constantly shaking.

• The Filter: The robot uses a special filter (like a noise-canceling headphone) to ignore the rapid shaking of the wings and focus only on the "big picture" direction it wants to go.
• The Control: It uses two main levers to fly:
  1. Angle Offset: Tilting the whole wing up or down to climb or dive.
  2. Stroke Timing (STAR): Changing the rhythm of the flaps to turn left or right.

5. Why This Matters

Why do we care about a 26-gram robot that looks like a butterfly?

• Stealth and Safety: Because it's light and flexible, if it bumps into a tree or a person, it won't break or hurt anyone. It's like a soft pillow compared to a hard drone.
• Confined Spaces: It can fly through tiny gaps, dense forests, or inside collapsed buildings where big, rigid drones can't go.
• Nature's Blueprint: By copying the butterfly, the engineers discovered that you don't need to be stiff to be stable. Sometimes, being flexible and embracing the wobble is the key to agility.

The Bottom Line

AirPulse is a proof of concept that we can build robots that fly like nature intended: messy, flexible, and full of life. It proves that you don't need a tail or a rigid frame to fly; you just need the right rhythm and a brain that knows how to dance with the wind. It opens the door for a new generation of robots that can quietly monitor ecosystems or inspect tight spaces without the danger of traditional drones.

Technical Summary

Here is a detailed technical summary of the paper "A 26-Gram Butterfly-Inspired Robot Achieving Autonomous Tailless Flight."

1. Problem Statement

Flapping-wing Micro Aerial Vehicles (FWMAVs) have largely focused on insect-scale (<1g) or bird/bat-inspired designs. However, the flight regime of butterflies remains underexplored in autonomous robotics. Butterflies present a unique aerodynamic challenge:

• Low-Frequency, High-Amplitude Flight: Unlike high-frequency insects (>150 Hz), butterflies flap at low frequencies (~10 Hz) with large amplitudes.
• Tailless and Underactuated: They lack tails or auxiliary control surfaces, relying on two wings for 6-DOF control.
• Strong Fluid-Structure-Inertial Coupling: Their compliant, low-aspect-ratio wings induce significant body undulations and inertial fluctuations. The variation in the center of gravity (CG) and moment of inertia during a wingbeat creates strong internal reaction torques that destabilize the body.
• Control Gap: Existing tailless FWMAVs typically use high-frequency kinematics to minimize these perturbations. There is a lack of hardware-validated models that can achieve closed-loop, onboard controlled flight at the butterfly scale (20–30g) while replicating these complex, oscillatory dynamics.

2. Methodology

The authors developed AirPulse, a 26g butterfly-inspired robot, and a novel control framework to address the challenges of tailless, low-frequency flight.

A. Biomimetic Hardware Design

• Mass & Scale: Total flight-ready mass is ~26g (dry mass ~24g), with a wingspan of 60mm. It is the lightest reported butterfly-inspired FWMAV with full onboard sensing and control.
• Wing Structure:
  • Material: Ultra-thin PET membrane (12.5 µm) reinforced with carbon-fiber rods arranged in a venation-inspired layout.
  • Stiffness Gradient: The design mimics biological wings with a graded stiffness distribution: rigid costal/basal regions for torque transfer and compliant distal regions for passive feathering (aerodynamic pitch rotation).
  • Geometry: Based on the Papilio demoleus (lime swallowtail), featuring a low aspect ratio (~3.28) and low wing loading (2.32 N/m²), the lowest among reported butterfly robots.
• Actuation: Two micro servos drive the forewings independently (anteromotoric actuation), mechanically coupling the fore- and hindwings. This setup replicates the phase lag and V-shaped dihedral found in nature.
• Avionics: A custom flight control board integrates a 9-axis IMU, barometer, and ESP32-S3 microcontroller, all co-located near the Center of Gravity (CG) to minimize inertial coupling.

B. Control Architecture & Algorithms

The system employs a hierarchical control architecture to manage the oscillatory dynamics:

1. State Estimation: A Madgwick filter fuses IMU data for real-time attitude estimation. Crucially, a Recursive Least Squares (RLS) estimator is used to extract the low-frequency mean of the pitch signal, filtering out the high-frequency body undulations caused by the wingbeat.
2. Attitude Controller: A PID controller computes errors between the estimated mean attitude and desired references, outputting modulation commands.
3. Central Pattern Generator (CPG): The controller translates commands into wing kinematics using a novel rhythm generator called Stroke Timing Asymmetry Rhythm (STAR).
   • STAR Mechanism: A single-parameter phase-domain generator that ensures continuous, smooth first-derivative kinematics. It modulates the upstroke/downstroke timing asymmetry linearly without causing phase discontinuities or actuator saturation.
   • Filtering: An IIR filter is applied to the reciprocal of the modulation function to prevent phase drift and ensure time consistency.

C. Control Effectiveness Mapping

The authors experimentally mapped flapping parameters (amplitude, frequency, stroke angle offset, and STAR timing) to force and torque generation:

• Amplitude/Frequency: Primarily control forward thrust and speed.
• Symmetric Angle Offset: Controls pitch (nose-up/down).
• Antisymmetric Angle Offset: Generates roll moments (though flight turns are primarily yaw-driven via adverse yaw).
• Symmetric STAR: Controls pitch (complementary to angle offset).
• Antisymmetric STAR: Generates roll moments for turning.

3. Key Contributions

1. First Onboard Controlled Tailless Flight: AirPulse is the first 26g, two-winged, tailless FWMAV to achieve fully autonomous, closed-loop flight with onboard sensing.
2. Novel Control Rhythm (STAR): Introduction of the STAR algorithm, which mathematically guarantees smooth, bounded, and monotonic stroke timing asymmetry, solving stability and continuity issues found in previous modulation methods.
3. Quantitative Control Mapping: The first systematic quantification of how flapping kinematics translate to forces and torques in a low-aspect-ratio, compliant wing system, filling a gap in understanding underactuated tailless flight.
4. Oscillatory Flight Control: Demonstration of a control strategy that explicitly accounts for and stabilizes the large body undulations inherent to low-frequency flapping, rather than trying to suppress them.

4. Results

• Flight Performance: AirPulse successfully performed stable free-flight maneuvers, including climbing and directed turning, without external tethers or tails.
• Dynamic Behavior: The robot exhibited characteristic body undulations (CG shifts up to 3% of body length, inertia variations up to 2.5x per stroke), confirming the strong fluid-structure-inertial coupling.
• Efficiency:
  • Average power consumption: ~5.9 W.
  • Power loading: 4.38 g/W.
  • This is significantly more energy-efficient than comparable micro-quadrotors (which draw ~10.6 W in hover).
• Aerodynamic Regime: The robot operates at a Reynolds number of ~5,600 and a reduced frequency ($k > 1$), placing it firmly in the unsteady aerodynamic regime dominated by vortex mechanisms (LEV, wake capture), similar to real butterflies.
• Validation: Experiments confirmed that the RLS-filtered pitch signal allows for stable PID control despite the high-frequency oscillations, and that STAR modulation provides a robust mechanism for attitude adjustment.

5. Significance

• Biological Insight: AirPulse serves as a physical model to decode the complex dynamics of butterfly flight, validating hypotheses about how body undulation and compliant wings contribute to stability and maneuverability.
• Robotics Advancement: It establishes a new paradigm for tailless, underactuated aerial robots, proving that stable flight is achievable even in highly oscillatory dynamical regimes.
• Practical Applications: The robot's lightweight, compliant, and collision-resilient design makes it ideal for non-invasive ecological monitoring (e.g., pollination studies, insect interaction) and confined-space inspection where rigid drones are too bulky or dangerous.
• Future Directions: The work paves the way for integrating neuromorphic sensors (dynamic vision, olfaction) to create fully autonomous, bio-inspired agents capable of navigating complex, GPS-denied environments.