Musculoskeletal simulation of limb movement biomechanics in Drosophila melanogaster

Pembe Gizem ÖzdilEPFL, Switzerland [email protected] Chuanfang NingEPFL, Switzerland [email protected] Jasper S. PhelpsEPFL, Switzerland [email protected] Sibo Wang-ChenEPFL, Switzerland [email protected] Guy ElishaEPFL, Switzerland [email protected] Alexander BlankeUniversity of Bonn, · 2025

Read Musculoskeletal simulation of limb movement biomechanics in Drosophila melanogaster by Pembe Gizem ÖzdilEPFL, Switzerland [email protected] Chuanfang NingEPFL, Switzerland [email protected] Jasper S. PhelpsEPFL, Switzerland [email protected] Sibo Wang-ChenEPFL, Switzerland [email protected] Guy ElishaEPFL, Switzerland [email protected] Alexander BlankeUniversity of Bonn, (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2509-06426.txt.

Introduction

A fruit fly's leg, rebuilt muscle by muscle inside a computer, is presented in a 2025 paper led by Pembe Gizem Özdil with seven co-authors. Near-complete maps of the fly's nervous system, muscles, and exoskeleton exist, yet no anatomically grounded leg muscle model bridged motor neurons to joint motion. The team builds the first three-dimensional, data-driven musculoskeletal model of Drosophila legs, in two simulation platforms.

Hill-type muscles are fitted from high-resolution X-ray scans of several specimens, with a pipeline that optimizes unknown fly-specific parameters. Combined with pose estimates from behaving flies, the model replays walking and grooming with muscle actuation and predicts testable muscle synergies. Imitation learning tests show joint damping and stiffness speed policy learning.

Catalog record

CREATORPembe Gizem ÖzdilEPFL, Switzerland [email protected] Chuanfang NingEPFL, Switzerland [email protected] Jasper S. PhelpsEPFL, Switzerland [email protected] Sibo Wang-ChenEPFL, Switzerland [email protected] Guy ElishaEPFL, Switzerland [email protected] Alexander BlankeUniversity of Bonn,
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE48 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2509-06426

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Musculoskeletal simulation of limb movement biomechanics in Drosophila melanogaster Pembe Gizem ÖzdilEPFL, Switzerland [email protected] Chuanfang NingEPFL, Switzerland [email protected] Jasper S. PhelpsEPFL, Switzerland [email protected] Sibo Wang-ChenEPFL, Switzerland [email protected] Guy ElishaEPFL, Switzerland [email protected] Alexander BlankeUniversity of Bonn, Germany [email protected] Auke IjspeertEPFL, Switzerland [email protected] Pavan RamdyaEPFL, Switzerland [email protected] Abstract Computational models are critical to advance our understanding of how neural, biomechanical, and physical systems interact to orchestrate animal behaviors. Despite the availability of near-complete reconstructions of the Drosophila melanogaster central nervous system, musculature, and exoskeleton, anatomically and physically grounded models of fly leg muscles are still missing. These models provide an indispensable bridge between motor neuron activity and joint movements. Here, we introduce the first 3D, data-driven musculoskeletal model of Drosophila legs, implemented in both OpenSim and MuJoCo simulation environments. Our model incorporates a Hill-type muscle representation based on high-resolution X-ray scans from multiple fixed specimens. We present a pipeline for constructing muscle models using morphological imaging data and for optimizing unknown muscle parameters specific to the fly. We then combine our musculoskeletal models with detailed 3D pose estimation data from behaving flies to achieve muscle-actuated behavioral replay in OpenSim. Simulations of muscle activity across diverse walking and grooming behaviors predict coordinated muscle synergies that can be tested experimentally. Furthermore, by training imitation learning policies in MuJoCo, we test the effect of different passive joint properties on learning speed and find that damping and stiffness facilitate learning. Overall, our model enables the investigation of motor control in an experimentally tractable model organism, providing insights into how biomechanics contribute to generation of complex limb movements. Moreover, our model can be used to control embodied artificial agents to generate naturalistic and compliant locomotion in simulated environments. 1 Introduction Understanding how to coordinate multiple limbs with many degrees of freedom (DoFs) to accomplish diverse motor tasks is a long-standing challenge in both motor control neuroscience and robotics. Performing such systems-level investigations hinges upon detailed knowledge of musculoskeletal structures and their mechanical properties. Neuromechanical models, therefore, serve as a crucial tool enabling rapid experimentation and hypothesis testing under controlled conditions [ 1 , 2 , 3 , 4 , 5 ] . Muscle-driven systems are characterized by intrinsic compliance and redundancy which can simplify the control problems faced by artificial agents as they do for animals in the real world [ 6 , 7 , 8 ] . However, many existing models abstract away key elements of musculature and passive biomechanics [ 9 , 10 ] limiting their ability to capture the richness of neuromuscular coordination. This, in turn, limits their utility for understanding how the nervous system generates robust, adaptive behaviors. Overcoming this gap will require modeling a system that generates sufficiently complex limb-dependent movements using a tractably small neuromuscular controller. The fruit fly, Drosophila melanogaster , represents an ideal organism for this reverse-engineering challenge. Flies generate complex behaviors using a compact, well-characterized, and genetically accessible nervous system [ 11 , 12 ] . Despite substantial progress in anatomical (e.g., connectomic [ 13 , 14 ] ), neural [ 15 , 16 , 17 ] , and behavioral characterization [ 18 , 19 ] , recently published biomechanical models of Drosophila legs [ 20 , 10 , 21 ] lack the anatomical and physiological accuracy required to model and investigate how motor networks coordinate muscles, limb kinematics, and behavior. Here, we present the first biologically detailed musculoskeletal model of Drosophila legs across two widely used physics engines, OpenSim and MuJoCo. Our model extends NeuroMechFly [ 10 , 21 ] , a whole-body biomechanical simulation of the adult fly, by incorporating anatomically detailed muscle representations and physiological properties. Specifically, we implement a Hill-type muscle model informed by anatomical data [ 22 , 23 , 24 ] to actuate seven DoFs in three leg joints. Using this model, we predict the relative contributions of individual muscles to joint movements generated by real behaving flies. Furthermore, we study the influence of passive joint properties on the successful imitation of animal behavior. Our work integrates anatomical, physiological, and behavioral data into a unified modeling framework, providing a critical foundation for future investigations of Drosophila neuromuscular coordination and behavioral control. This modeling framework ( Figure 1 ) can also accelerate the development of musculoskeletal models for other species, inform machine learning-based algorithms for embodied agents, and facilitate the control of bioinspired robots. Figure 1: Pipeline to develop Drosophila leg musculoskeletal models. (Left) Anatomical data from multiple flies were used to determine muscle attachment points and fiber paths, providing constraints for constructing the muscle model in OpenSim. (Right) 3D pose estimation data from behaving animals were then used to train the muscle-actuated agent to recapitulate detailed kinematics in OpenSim and MuJoCo. (Middle) The resulting muscle model informed by both leg anatomy and joint kinematics. 2 Related work 2.1 Musculoskeletal models of animals Musculoskeletal models have emerged as powerful tools for investigating animal biomechanics. Open-source simulation platforms such as OpenSim [ 4 ] , MyoSuite [ 25 ] , and DART [ 26 ] , as well as commercial frameworks like HyFyDy [ 27 ] , have made it possible to test state-of-the-art learning algorithms on complex, high-dimensional systems. Use cases include the control of human arms for manipulation and legs for locomotion [ 28 , 29 ] . Although primarily developed to simulate human bodies [ 30 , 31 ] , these tools have enabled the creation of musculoskeletal models for a wide range of species, including rodents [ 32 , 33 , 34 , 35 ] , primates [ 36 , 37 ] , horses [ 38 ] , and birds like ostriches [ 39 , 40 ] . Researchers have also developed custom tools to study the interplay between neural circuits and biomechanics in worms [ 41 ] , insects [ 42 ] , and other animals [ 43 ] . Recently, the fruit fly, Drosophila melanogaster , has become an increasingly prominent animal model for musculoskeletal simulations. The first anatomically realistic fly body model, NeuroMechFly, used micro-computed tomography (micro-CT) scans to construct a morphologically accurate 3D fly simulation. In its earliest incarnation, simplified antagonistic spring-damper muscles were used in the legs of NeuroMechFly to simulate locomotor dynamics [ 21 ] . More recent efforts in this [ 10 ] and other [ 20 ] fly body models have enhanced simulations by adding, for example, multimodal sensorimotor transformations to accomplish a wider range of behavioral tasks. However, these models still rely on abstract —position- or torque-based— joint controllers and lack realistic muscle actuation. We address this gap in biological fidelity in our work by explicitly modeling individual fly leg muscles based on detailed anatomical data. 2.2 Motor control and muscle synergies Coordinating the activities of many muscles is a fundamental challenge in motor control. It has been proposed that the nervous system simplifies this complex problem by using muscle synergies—groups of muscles that are co-active, acting as functional units [ 44 , 45 ] . By combining a limited number of these synergies, the brain can more efficiently produce diverse movements without needing to control each muscle independently [ 44 , 46 ] . Muscle synergies have been identified and analyzed in real experimental data [ 47 , 48 ,

Read the complete text