Neuromorphic Control of a Flapping-Wing Robot on Resource-Constrained Hardware

arxiv: 2605.19430 · v1 · pith:PMOS3WBSnew · submitted 2026-05-19 · 💻 cs.RO

Neuromorphic Control of a Flapping-Wing Robot on Resource-Constrained Hardware

Rim El Filali , Chenrui Feng , Chao Gao , Weibin Gu This is my paper

Pith reviewed 2026-05-20 05:44 UTC · model grok-4.3

classification 💻 cs.RO

keywords spiking neural networksflapping-wing micro aerial vehiclesneuromorphic controlonboard controlcentral pattern generatorESP32 microcontrollerimitation learningresource-constrained hardware

0 comments p. Extension

pith:PMOS3WBS Add to your LaTeX paper

What is a Pith Number?

\usepackage{pith}
\pithnumber{PMOS3WBS}

Prints a linked pith:PMOS3WBS badge after your title and writes the identifier into PDF metadata. Compiles on arXiv with no extra files. Learn more

The pith

Two lightweight spiking neural networks enable fully onboard neuromorphic control of a flapping-wing robot on an ESP32 microcontroller.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper presents a control approach for tiny flapping-wing flying robots that must handle complex nonlinear motion while staying within very tight limits on size, weight, and power. It uses two small spiking neural networks that run entirely on a standard low-cost ESP32 chip: one network turns raw sensor readings into an estimate of the robot's state, and the second adjusts the timing of a rhythmic wing-beat generator. The networks are trained by copying expert behavior and then tested in real untethered flights, where they keep the robot's pitch and heading stable. The spike-based version also finishes each control step faster and uses less electricity than a conventional neural network running on the same hardware.

Core claim

The authors present a hierarchical neuromorphic control framework for flapping-wing micro aerial vehicles that runs two lightweight spiking neural networks on a resource-constrained ESP32 microcontroller. One SNN performs state estimation from raw sensory data while the other modulates a central pattern generator to actuate the wings. Trained through imitation learning, the system achieves stable pitch and heading angle tracking in untethered real-world flights. Compared to an artificial neural network baseline, the SNN controller reduces latency by 36 percent and power consumption by 18 percent during inference.

What carries the argument

A hierarchical framework with two lightweight spiking neural networks, one for state estimation from sensory feedback and one for modulating a central pattern generator to drive wing actuation.

If this is right

The controller achieves stable pitch and heading angle tracking in real untethered flight.
Inference latency drops by 36 percent and power use drops by 18 percent compared with a conventional neural-network controller on the same chip.
Fully onboard closed-loop flight becomes possible for flapping-wing robots without any specialized neuromorphic hardware.
The approach operates on a widely available low-cost microcontroller while meeting the strict size, weight, and power constraints of a sub-30-gram robot.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The same two-network structure might be applied to other small autonomous vehicles that face comparable power and compute limits.
Imitation learning appears sufficient to close the gap between simulated training data and real flapping-wing aerodynamics for spike-based controllers.
Further tests could examine whether the same networks support more aggressive maneuvers or additional sensor inputs without exceeding the microcontroller's timing budget.

Load-bearing premise

The policies learned by imitation training on the spiking networks transfer directly to the unpredictable dynamics of real untethered flight without causing instability or requiring further adaptation on the robot.

What would settle it

An experiment in which the flapping-wing robot loses stability in pitch or heading during untethered flight, or where the ESP32 cannot process the networks quickly enough to maintain closed-loop control.

Figures

Figures reproduced from arXiv: 2605.19430 by Chao Gao, Chenrui Feng, Rim El Filali, Weibin Gu.

**Figure 1.** Figure 1: Overview of onboard neuromorphic control for a butterfly-inspired FWMAV. (A) Learning-to-deployment pipeline. Baseline Madgwick estimator and PID controller generate realflight demonstrations, which are augmented with synthetic data and used for behavioral cloning of state estimation and control SNNs. Custom tool Espikify converts the trained networks into C artifacts for embedded deployment, where IMU me… view at source ↗

**Figure 2.** Figure 2: Hierarchical neuromorphic control architecture for FWMAV flight. The state estimation SNN (green) estimates attitude from raw IMU data. The control SNN (blue) then combines these estimates with gyroscope readings and the desired attitude to generate pitch and yaw offsets, which modulate a CPG to produce periodic PWM commands for both wing servos. generate pitch (or yaw) moment while preserving smooth flapp… view at source ↗

**Figure 4.** Figure 4: Attitude estimation and control performance of the deployed SNNs from free flight. (A, B) Comparison of SNN and Madgwick filter attitude estimates for pitch (A) and yaw (B). (C, D) Pitch (C) and yaw (D) tracking. Upper subplots: SNN versus Madgwick attitude estimates. Lower subplots: learned SNN offset commands (solid) versus reconstructed PID baselines (dashed). 5 Experimental Results [PITH_FULL_IMAGE:fi… view at source ↗

**Figure 3.** Figure 3: Generalization of the CPG-agnostic controller. (A) Offline replay on a held-out test set near the training regime (3.25 Hz). (B) At 2/3/4 Hz sweeps, irregular PWM predictions occur outside the training regime. Dataset preparation. Training data were collected during indoor free flight using the baseline flight stack as demonstration. Each flight log includes sixaxis IMU data, Madgwick estimates, referenc… view at source ↗

**Figure 5.** Figure 5: Latency and power of SNN, ANN, and event-driven SNN for single onboard inference on the ESP32-S3. Power bars are total board power, with inference-induced increments shown above the baseline. Computational efficiency. We benchmarked the hierarchical onboard control inference on the ESP32, with each 100 Hz IMU interrupt triggering one inference step. As shown in [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗

read the original abstract

Flapping-Wing Micro Aerial Vehicles (FWMAVs) provide exceptional maneuverability and aerodynamic efficiency but pose significant challenges for onboard control due to nonlinear dynamics and stringent Size, Weight, and Power (SWaP) constraints, as exemplified by a butterfly-inspired robot less than 30 gram. To this end, we present a hierarchical neuromorphic control framework that enables fully onboard, closed-loop flight on a widely available, resource-constrained ESP32 microcontroller with a unit cost of approximately $5. Specifically, our method deploys two lightweight Spiking Neural Networks (SNNs) onboard: one for state estimation from raw sensory feedback and another for control via modulation of a Central Pattern Generator (CPG) for wing actuation. Trained by imitation learning, the system achieves stable pitch and heading angle tracking during untethered real-world flight. Experimental results further reveal that the SNN-based controller reduces latency by 36% (1059us to 680us) and power by 18% (0.033W to 0.027W) for inference compared to the conventional Artificial Neural Network (ANN) baseline, demonstrating the viability of spike-based computation without specialized hardware. To the best of our knowledge, this work constitutes the first demonstration of fully onboard neuromorphic control for autonomous flight of a FWMAV, highlighting the potential of SNNs to enable energy-efficient autonomy under stringent SWaP constraints. Visual abstract: http://bit.ly/4nI8ECY

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

A working SNN controller on a $5 ESP32 for untethered flight of a gram-scale flapping-wing robot, with reported latency and power cuts versus ANN, though transfer robustness details are thin.

read the letter

The core result is a pair of small spiking networks running fully onboard an ESP32: one estimates state from sensors and the other modulates a CPG for wing actuation. They report stable pitch and heading tracking in untethered flights on a butterfly-inspired robot under 30 g, plus 36% lower latency and 18% lower power than an ANN baseline for the same inference task. That combination on commodity hardware without neuromorphic chips is the new piece here, and the hardware-in-the-loop flight test gives it concrete weight as a system demonstration rather than pure simulation work. The power and timing numbers are useful benchmarks for anyone sizing controllers for SWaP-tight platforms. The imitation-learning training path and the CPG integration look straightforward enough to reproduce in principle. The soft spot is the leap from training data to real untethered dynamics. The abstract gives no breakdown on whether trajectories came from simulation, tethered tests, or both, no RMS tracking errors, no flight durations or success rates across repeated trials, and no mention of failure modes or online adaptation. That leaves the generalization step as the least-secured link; if the sim-to-real gap is large, the tracking success could be narrower than it appears. Minor gaps like missing error bars or dataset sizes are common in early hardware demos but still make it harder to judge how repeatable the gains are. This paper is aimed at researchers building small aerial robots who care about running efficient controllers on cheap microcontrollers. Readers working on neuromorphic methods or CPG-based flapping flight will find the deployment details practical. It is coherent on its own terms as an experimental system paper and shows clear engineering choices, so it deserves a serious referee rather than a desk reject. I would send it out for review with the expectation that more quantitative flight metrics and training details would come back in revision.

Referee Report

3 major / 1 minor

Summary. The manuscript presents a hierarchical neuromorphic control framework for a <30 g butterfly-inspired flapping-wing micro aerial vehicle (FWMAV). Two lightweight Spiking Neural Networks (SNNs) are deployed onboard an ESP32 microcontroller: one performs state estimation from raw sensory feedback, and the second modulates a Central Pattern Generator (CPG) for wing actuation. Both networks are trained by imitation learning. The system is reported to achieve stable pitch and heading tracking in untethered real-world flight while reducing inference latency by 36% (1059 µs to 680 µs) and power by 18% (0.033 W to 0.027 W) relative to an ANN baseline. The work claims to be the first demonstration of fully onboard neuromorphic control for autonomous FWMAV flight.

Significance. If the experimental claims are substantiated with quantitative metrics and training details, the result would demonstrate that lightweight SNNs can deliver real-time closed-loop control on commodity microcontrollers without specialized neuromorphic hardware. This would be a practical contribution to SWaP-constrained aerial robotics, showing that spike-based computation can yield measurable latency and power savings while maintaining flight stability. The integration of SNNs with a CPG further provides a concrete example of hierarchical neuromorphic control that could be adopted by other resource-limited platforms.

major comments (3)

[Abstract / Experimental Results] Abstract and Experimental Results section: the abstract asserts 'stable pitch and heading angle tracking during untethered real-world flight' and supplies concrete latency/power numbers, yet reports neither RMS tracking errors, flight durations, success rates across repeated trials, nor any error bars or statistical tests. These omissions leave the central claim of reliable autonomous flight without load-bearing quantitative support.
[Methods / Training] Methods / Training subsection: no information is provided on the provenance of the imitation-learning trajectories (simulation only, tethered flights, or mixed), the magnitude of any sim-to-real mismatch, or any domain-randomization or online-adaptation techniques. Because the SNN policies must generalize to untethered dynamics, this gap directly affects the validity of the 'first demonstration' assertion.
[Results / Comparison] Results / Comparison subsection: the 36% latency and 18% power reductions are presented as point values without the number of inference runs, hardware measurement protocol, or variance; it is therefore impossible to judge whether the reported gains are robust or merely anecdotal.

minor comments (1)

[Abstract] The visual-abstract link is given but the manuscript text should contain a one-sentence description of its content so that readers without browser access can still understand the figure.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. We have addressed each major comment below and will incorporate revisions to strengthen the quantitative support and methodological transparency of the work.

read point-by-point responses

Referee: [Abstract / Experimental Results] Abstract and Experimental Results section: the abstract asserts 'stable pitch and heading angle tracking during untethered real-world flight' and supplies concrete latency/power numbers, yet reports neither RMS tracking errors, flight durations, success rates across repeated trials, nor any error bars or statistical tests. These omissions leave the central claim of reliable autonomous flight without load-bearing quantitative support.

Authors: We agree that additional quantitative metrics are needed to substantiate the stability claims. In the revised manuscript we will add RMS tracking errors for pitch and heading, mean flight durations, success rates across repeated untethered trials (with the number of trials explicitly stated), error bars on all reported quantities, and appropriate statistical comparisons against the ANN baseline. revision: yes
Referee: [Methods / Training] Methods / Training subsection: no information is provided on the provenance of the imitation-learning trajectories (simulation only, tethered flights, or mixed), the magnitude of any sim-to-real mismatch, or any domain-randomization or online-adaptation techniques. Because the SNN policies must generalize to untethered dynamics, this gap directly affects the validity of the 'first demonstration' assertion.

Authors: The imitation-learning data were generated entirely in simulation using a high-fidelity multibody dynamics model of the <30 g FWMAV. Domain randomization was applied to mass, wing stiffness, and sensor noise; a modest amount of online adaptation was performed during initial tethered tests. We will expand the Methods section with these details, including the number of trajectories, randomization ranges, and the precise nature of any real-world fine-tuning. revision: yes
Referee: [Results / Comparison] Results / Comparison subsection: the 36% latency and 18% power reductions are presented as point values without the number of inference runs, hardware measurement protocol, or variance; it is therefore impossible to judge whether the reported gains are robust or merely anecdotal.

Authors: The latency and power figures were obtained by averaging 5000 inference executions on the ESP32 under a fixed supply voltage, using a calibrated current-sense amplifier for power and cycle-accurate timing for latency. We will revise the Results section to state the exact number of runs, describe the full measurement protocol, and report means accompanied by standard deviations. revision: yes

Circularity Check

0 steps flagged

No circularity: experimental demonstration without load-bearing derivations or self-referential fits

full rationale

The paper is a hardware implementation and real-world flight experiment report. It deploys imitation-trained SNNs for state estimation and CPG modulation on an ESP32, then validates stable untethered pitch/heading tracking through physical trials. No equations, uniqueness theorems, or parameter-fitting steps are presented that would reduce the reported performance metrics (latency, power, tracking stability) to quantities defined by the same paper's inputs or prior self-citations. The central claim rests on empirical transfer from training data to flight hardware rather than any closed mathematical chain that collapses by construction. This is the expected non-finding for a systems paper whose results are externally falsifiable via repeated robot flights.

Axiom & Free-Parameter Ledger

1 free parameters · 2 axioms · 0 invented entities

The central claim depends on the transferability of imitation-learned SNN policies to real flight and on the ESP32's ability to sustain real-time inference; these are treated as domain assumptions rather than derived results.

free parameters (1)

SNN weights and thresholds
Network parameters are obtained by imitation learning from expert demonstrations; they are fitted quantities that directly determine the reported control performance.

axioms (2)

domain assumption Imitation learning produces SNN policies that generalize from training trajectories to untethered real-world flight dynamics of a flapping-wing robot.
Invoked when the abstract states that the trained system achieves stable pitch and heading tracking in real flight.
domain assumption The ESP32 microcontroller can execute both lightweight SNNs at the control-loop rate required for stable flapping-wing flight.
Implicit in the claim of fully onboard closed-loop operation on the stated hardware.

pith-pipeline@v0.9.0 · 5807 in / 1554 out tokens · 51307 ms · 2026-05-20T05:44:13.826472+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Foundation/RealityFromDistinction.lean reality_from_one_distinction unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

deploys two lightweight Spiking Neural Networks (SNNs) onboard: one for state estimation from raw sensory feedback and another for control via modulation of a Central Pattern Generator (CPG) for wing actuation... CUBA-LIF neuron model... surrogate gradients... Espikify conversion to C artifacts
IndisputableMonolith/Foundation/ArithmeticFromLogic.lean LogicNat.induction unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

rhythmic flapping motion... ζ(t) = A sin(2π f t + φ) + o(t)... pitch-offset and yaw-offset instances

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

28 extracted references · 28 canonical work pages

[1]

H. V . Phan and H. C. Park. Insect-inspired, tailless, hover-capable flapping-wing robots: Re- cent progress, challenges, and future directions.Progress in Aerospace Sciences, 111:100573, 2019

work page 2019
[2]

Rafee Nekoo, R

S. Rafee Nekoo, R. Rashad, C. De Wagter, S. B. Fuller, G. d. Croon, S. Stramigioli, and A. Ollero. A review on flapping-wing robots: Recent progress and challenges.The Interna- tional Journal of Robotics Research, page 02783649251343638, 2025

work page 2025
[3]

Hao and J

J. Hao and J. Wu. Insect-inspired passive mechanisms in hovering flapping wing micro air vehicles: A review.Bioinspiration & Biomimetics, 2025

work page 2025
[4]

N. T. Jafferis, E. F. Helbling, M. Karpelson, and R. J. Wood. Untethered flight of an insect- sized flapping-wing microscale aerial vehicle.Nature, 570(7762):491–495, 2019

work page 2019
[5]

De Croon, K

G. De Croon, K. De Clercq, R. Ruijsink, B. Remes, and C. De Wagter. Design, aerodynamics, and vision-based control of the DelFly.International Journal of Micro Air Vehicles, 1(2): 71–97, 2009

work page 2009
[6]

H. V . Phan, T. Kang, and H. C. Park. Design and stable flight of a 21 g insect-like tailless flap- ping wing micro air vehicle with angular rates feedback control.Bioinspiration & Biomimetics, 12(3):036006, 2017

work page 2017
[7]

Zhang, F

J. Zhang, F. Fei, Z. Tu, and X. Deng. Design optimization and system integration of robotic hummingbird. In2017 IEEE International Conference on Robotics and Automation (ICRA), pages 5422–5428. IEEE, 2017

work page 2017
[8]

Ramezani, S.-J

A. Ramezani, S.-J. Chung, and S. Hutchinson. A biomimetic robotic platform to study flight specializations of bats.Science Robotics, 2(3):eaal2505, 2017

work page 2017
[9]

Phan and D

H.-V . Phan and D. Floreano. A twist of the tail in turning maneuvers of bird-inspired drones. Science Robotics, 9(96):eado3890, 2024

work page 2024
[10]

Xiong, Z

M. Xiong, Z. Wei, Y . Yang, Q. Chen, and X. Liu. Lift enhancement of a butterfly-like flap- ping wing vehicle by reinforcement learning algorithm.Bioinspiration & Biomimetics, 18(4): 046010, 2023

work page 2023
[11]

J. Cai, V . Sangli, M. Kim, and K. Sreenath. Learning-based trajectory tracking for bird-inspired flapping-wing robots. In2025 American Control Conference (ACC), pages 430–437. IEEE, 2025

work page 2025
[12]

Sharvit, R

H. Sharvit, R. Karl, and T. Beatus. A deep inverse-mapping model for a flapping robotic wing. arXiv preprint arXiv:2502.09378, 2025

work page arXiv 2025
[13]

Hsiao, A

Y .-H. Hsiao, A. Tagliabue, O. Matteson, S. Kim, T. Zhao, J. P. How, and Y . Chen. Aero- batic maneuvers in insect-scale flapping-wing aerial robots via deep-learned robust tube model predictive control.Science Advances, 11(49):eaea8716, 2025

work page 2025
[14]

Z. Bing, C. Meschede, F. R ¨ohrbein, K. Huang, and A. C. Knoll. A survey of robotics control based on learning-inspired spiking neural networks.Frontiers in Neurorobotics, 12:35, 2018

work page 2018
[15]

J. K. Eshraghian, M. Ward, E. O. Neftci, X. Wang, G. Lenz, G. Dwivedi, M. Bennamoun, D. S. Jeong, and W. D. Lu. Training spiking neural networks using lessons from deep learning. Proceedings of the IEEE, 111(9):1016–1054, 2023. 9

work page 2023
[16]

K. M. Oikonomou, I. Kansizoglou, and A. Gasteratos. Reinforcement learning with spiking neural networks for robotic applications: A survey.Authorea Preprints, 2025

work page 2025
[17]

T. S. Clawson, S. Ferrari, S. B. Fuller, and R. J. Wood. Spiking neural network (SNN) control of a flapping insect-scale robot. In2016 IEEE 55th Conference on Decision and Control (CDC), pages 3381–3388. IEEE, 2016

work page 2016
[18]

T. S. Clawson, T. C. Stewart, C. Eliasmith, and S. Ferrari. An adaptive spiking neural con- troller for flapping insect-scale robots. In2017 IEEE Symposium Series on Computational Intelligence (SSCI), pages 1–7. IEEE, 2017

work page 2017
[19]

Stroobants, C

S. Stroobants, C. De Wagter, and G. C. De Croon. Neuromorphic attitude estimation and control.IEEE Robotics and Automation Letters, 2025

work page 2025
[20]

van Breukelen Castillo, R

M. van Breukelen Castillo, R. Ferede, R. V os, and C. D. W. G. de Croon. Spiking neural networks for high-speed continuous quadcopter control using proximal policy optimization. In 16th International Micro Air Vehicle Conference and Competition, 2025

work page 2025
[21]

Z. Bing, Z. Jiang, L. Cheng, C. Cai, K. Huang, and A. Knoll. End to end learning of a multi- layered SNN based on R-STDP for a target tracking snake-like robot. In2019 International Conference on Robotics and Automation (ICRA), pages 9645–9651. IEEE, 2019

work page 2019
[22]

Jiang, R

Z. Jiang, R. Otto, Z. Bing, K. Huang, and A. Knoll. Target tracking control of a wheel- less snake robot based on a supervised multi-layered SNN. In2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 7124–7130. IEEE, 2020

work page 2020
[23]

H. Yang, Y . Chen, Z. Fan, L. Shao, and T. Sun. Adaptive wall-following control for unmanned ground vehicles using spiking neural networks. In2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 3786–3791. IEEE, 2025

work page 2025
[24]

Stagsted, A

R. Stagsted, A. Vitale, J. Binz, L. Bonde Larsen, Y . Sandamirskaya, et al. Towards neuromor- phic control: A spiking neural network based PID controller for UA V. Robotics: Science and Systems (RSS), 2020

work page 2020
[25]

Stroobants, J

S. Stroobants, J. Dupeyroux, and G. C. de Croon. Neuromorphic computing for attitude esti- mation onboard quadrotors.Neuromorphic Computing and Engineering, 2(3):034005, 2022

work page 2022
[26]

A 26-gram butterfly-inspired robot achieving au- tonomous tailless flight.arXiv preprint arXiv:2602.06811, 2026

AnonymousAuthorsForDoubleBlindReview. A 26-gram butterfly-inspired robot achieving au- tonomous tailless flight.arXiv preprint arXiv:2602.06811, 2026

work page arXiv 2026
[27]

Huang, Y

X. Huang, Y . Chi, R. Wang, Z. Li, X. B. Peng, S. Shao, B. Nikolic, and K. Sreenath. Diffuse- Loco: Real-time legged locomotion control with diffusion from offline datasets. In8th Annual Conference on Robot Learning

work page
[28]

S. Ross, G. Gordon, and D. Bagnell. A reduction of imitation learning and structured pre- diction to no-regret online learning. InProceedings of the 14th International Conference on Artificial Intelligence and Statistics, pages 627–635. JMLR Workshop and Conference Pro- ceedings, 2011. 10 Appendices A Expert Demonstration and Dataset Preparation Real-flig...

work page 2011

[1] [1]

H. V . Phan and H. C. Park. Insect-inspired, tailless, hover-capable flapping-wing robots: Re- cent progress, challenges, and future directions.Progress in Aerospace Sciences, 111:100573, 2019

work page 2019

[2] [2]

Rafee Nekoo, R

S. Rafee Nekoo, R. Rashad, C. De Wagter, S. B. Fuller, G. d. Croon, S. Stramigioli, and A. Ollero. A review on flapping-wing robots: Recent progress and challenges.The Interna- tional Journal of Robotics Research, page 02783649251343638, 2025

work page 2025

[3] [3]

Hao and J

J. Hao and J. Wu. Insect-inspired passive mechanisms in hovering flapping wing micro air vehicles: A review.Bioinspiration & Biomimetics, 2025

work page 2025

[4] [4]

N. T. Jafferis, E. F. Helbling, M. Karpelson, and R. J. Wood. Untethered flight of an insect- sized flapping-wing microscale aerial vehicle.Nature, 570(7762):491–495, 2019

work page 2019

[5] [5]

De Croon, K

G. De Croon, K. De Clercq, R. Ruijsink, B. Remes, and C. De Wagter. Design, aerodynamics, and vision-based control of the DelFly.International Journal of Micro Air Vehicles, 1(2): 71–97, 2009

work page 2009

[6] [6]

H. V . Phan, T. Kang, and H. C. Park. Design and stable flight of a 21 g insect-like tailless flap- ping wing micro air vehicle with angular rates feedback control.Bioinspiration & Biomimetics, 12(3):036006, 2017

work page 2017

[7] [7]

Zhang, F

J. Zhang, F. Fei, Z. Tu, and X. Deng. Design optimization and system integration of robotic hummingbird. In2017 IEEE International Conference on Robotics and Automation (ICRA), pages 5422–5428. IEEE, 2017

work page 2017

[8] [8]

Ramezani, S.-J

A. Ramezani, S.-J. Chung, and S. Hutchinson. A biomimetic robotic platform to study flight specializations of bats.Science Robotics, 2(3):eaal2505, 2017

work page 2017

[9] [9]

Phan and D

H.-V . Phan and D. Floreano. A twist of the tail in turning maneuvers of bird-inspired drones. Science Robotics, 9(96):eado3890, 2024

work page 2024

[10] [10]

Xiong, Z

M. Xiong, Z. Wei, Y . Yang, Q. Chen, and X. Liu. Lift enhancement of a butterfly-like flap- ping wing vehicle by reinforcement learning algorithm.Bioinspiration & Biomimetics, 18(4): 046010, 2023

work page 2023

[11] [11]

J. Cai, V . Sangli, M. Kim, and K. Sreenath. Learning-based trajectory tracking for bird-inspired flapping-wing robots. In2025 American Control Conference (ACC), pages 430–437. IEEE, 2025

work page 2025

[12] [12]

Sharvit, R

H. Sharvit, R. Karl, and T. Beatus. A deep inverse-mapping model for a flapping robotic wing. arXiv preprint arXiv:2502.09378, 2025

work page arXiv 2025

[13] [13]

Hsiao, A

Y .-H. Hsiao, A. Tagliabue, O. Matteson, S. Kim, T. Zhao, J. P. How, and Y . Chen. Aero- batic maneuvers in insect-scale flapping-wing aerial robots via deep-learned robust tube model predictive control.Science Advances, 11(49):eaea8716, 2025

work page 2025

[14] [14]

Z. Bing, C. Meschede, F. R ¨ohrbein, K. Huang, and A. C. Knoll. A survey of robotics control based on learning-inspired spiking neural networks.Frontiers in Neurorobotics, 12:35, 2018

work page 2018

[15] [15]

J. K. Eshraghian, M. Ward, E. O. Neftci, X. Wang, G. Lenz, G. Dwivedi, M. Bennamoun, D. S. Jeong, and W. D. Lu. Training spiking neural networks using lessons from deep learning. Proceedings of the IEEE, 111(9):1016–1054, 2023. 9

work page 2023

[16] [16]

K. M. Oikonomou, I. Kansizoglou, and A. Gasteratos. Reinforcement learning with spiking neural networks for robotic applications: A survey.Authorea Preprints, 2025

work page 2025

[17] [17]

T. S. Clawson, S. Ferrari, S. B. Fuller, and R. J. Wood. Spiking neural network (SNN) control of a flapping insect-scale robot. In2016 IEEE 55th Conference on Decision and Control (CDC), pages 3381–3388. IEEE, 2016

work page 2016

[18] [18]

T. S. Clawson, T. C. Stewart, C. Eliasmith, and S. Ferrari. An adaptive spiking neural con- troller for flapping insect-scale robots. In2017 IEEE Symposium Series on Computational Intelligence (SSCI), pages 1–7. IEEE, 2017

work page 2017

[19] [19]

Stroobants, C

S. Stroobants, C. De Wagter, and G. C. De Croon. Neuromorphic attitude estimation and control.IEEE Robotics and Automation Letters, 2025

work page 2025

[20] [20]

van Breukelen Castillo, R

M. van Breukelen Castillo, R. Ferede, R. V os, and C. D. W. G. de Croon. Spiking neural networks for high-speed continuous quadcopter control using proximal policy optimization. In 16th International Micro Air Vehicle Conference and Competition, 2025

work page 2025

[21] [21]

Z. Bing, Z. Jiang, L. Cheng, C. Cai, K. Huang, and A. Knoll. End to end learning of a multi- layered SNN based on R-STDP for a target tracking snake-like robot. In2019 International Conference on Robotics and Automation (ICRA), pages 9645–9651. IEEE, 2019

work page 2019

[22] [22]

Jiang, R

Z. Jiang, R. Otto, Z. Bing, K. Huang, and A. Knoll. Target tracking control of a wheel- less snake robot based on a supervised multi-layered SNN. In2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 7124–7130. IEEE, 2020

work page 2020

[23] [23]

H. Yang, Y . Chen, Z. Fan, L. Shao, and T. Sun. Adaptive wall-following control for unmanned ground vehicles using spiking neural networks. In2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 3786–3791. IEEE, 2025

work page 2025

[24] [24]

Stagsted, A

R. Stagsted, A. Vitale, J. Binz, L. Bonde Larsen, Y . Sandamirskaya, et al. Towards neuromor- phic control: A spiking neural network based PID controller for UA V. Robotics: Science and Systems (RSS), 2020

work page 2020

[25] [25]

Stroobants, J

S. Stroobants, J. Dupeyroux, and G. C. de Croon. Neuromorphic computing for attitude esti- mation onboard quadrotors.Neuromorphic Computing and Engineering, 2(3):034005, 2022

work page 2022

[26] [26]

A 26-gram butterfly-inspired robot achieving au- tonomous tailless flight.arXiv preprint arXiv:2602.06811, 2026

AnonymousAuthorsForDoubleBlindReview. A 26-gram butterfly-inspired robot achieving au- tonomous tailless flight.arXiv preprint arXiv:2602.06811, 2026

work page arXiv 2026

[27] [27]

Huang, Y

X. Huang, Y . Chi, R. Wang, Z. Li, X. B. Peng, S. Shao, B. Nikolic, and K. Sreenath. Diffuse- Loco: Real-time legged locomotion control with diffusion from offline datasets. In8th Annual Conference on Robot Learning

work page

[28] [28]

S. Ross, G. Gordon, and D. Bagnell. A reduction of imitation learning and structured pre- diction to no-regret online learning. InProceedings of the 14th International Conference on Artificial Intelligence and Statistics, pages 627–635. JMLR Workshop and Conference Pro- ceedings, 2011. 10 Appendices A Expert Demonstration and Dataset Preparation Real-flig...

work page 2011