FARMS: Framework for Animal and Robot Modeling and Simulation

The paper introduces FARMS, an open-source, interdisciplinary Python framework that unifies 3D modeling, simulation, and analysis tools to facilitate the construction and study of complex neuromechanical systems for both animals and robots across diverse environments.

The Gist

Imagine trying to understand how a salamander swims in a pond and then immediately hops onto a rock to walk. It looks effortless, but underneath that smooth motion is a chaotic dance of muscles firing, nerves sending signals, bones pushing against water, and friction against the ground.

For a long time, scientists had to build separate "labs" to study each part of this dance. One team built robots to test mechanics; another built computer models to test brains; a third used complex math to predict movement. They rarely talked to each other, and switching between these different tools was like trying to play a video game where you have to switch consoles every time you want to change the level.

Enter FARMS.

Think of FARMS (Framework for Animal and Robot Modeling and Simulation) as the "Universal Translator and Lego Kit" for scientists who study movement. It is a free, open-source software tool that lets researchers build, test, and analyze how animals and robots move in a single, unified workspace.

Here is how it works, using some everyday analogies:

1. The "All-in-One" Workshop

Imagine you are building a model car. Usually, you might need one box for the wheels, another for the engine, and a third for the paint. If you want to change the engine, you have to take the whole thing apart and start over.

FARMS is like a super-workbench where the engine, the wheels, the paint, and the driver's instructions are all in one place.

• The Body (The Chassis): You can build a digital skeleton for a robot or a real animal (like a mouse or a salamander).
• The Brain (The Driver): You can program a "brain" for it. This could be a simple computer algorithm or a complex biological neural network (like a real spinal cord).
• The World (The Track): You can drop your creation into a virtual world. It could be a dry floor, a muddy swamp, or a flowing river.

2. The "Magic Glue" (Interoperability)

One of the biggest headaches in science is that different software programs speak different languages. A program that simulates muscles might not talk to the program that simulates water.

FARMS acts as the magic glue. It connects different open-source tools (like physics engines that calculate how things bounce, and neural networks that calculate how brains think) so they work together seamlessly.

• Analogy: It's like having a universal power strip. Whether you plug in a toaster (muscle model), a blender (physics engine), or a lamp (neural network), they all get power and work together without needing a different adapter for each one.

3. What Can You Do With It? (The Demonstrations)

The paper shows off what this "magic kit" can do with some cool examples:

• The Shape-Shifter (Robots): They simulated robots that look like eels and salamanders. These robots could swim in a virtual pool and, the moment they touched a ramp, their "brain" would switch gears, and they would start walking on land. FARMS handled the tricky physics of water resistance and dry friction automatically.
• The Blind Swimmer (Real Animals): They built a digital twin of a real salamander. They put it in a river with a current. The digital salamander didn't "know" it was in water; it just reacted to the water pushing against its skin. It switched from walking to swimming and back again purely based on physical contact. This helps scientists understand how real animals do it without needing to hurt or stress a live animal.
• The Snake in the Pegs: They simulated a snake trying to slither through a forest of vertical pegs. The snake used the pegs to push itself forward. This showed how a simple "wiggling" brain pattern could solve a complex obstacle course just by interacting with the environment.
• The Mouse Builder: They took a real mouse, scanned its bones, and used FARMS to build a digital version with working muscles. They even used a special tool (a Blender plugin) to "paint" the muscles onto the bones, making it much easier to create these complex models than doing it by hand.

4. Why Does This Matter?

Before FARMS, if a scientist wanted to study how a robot learns to walk, they might spend months just setting up the software. With FARMS, they can focus on the science, not the setup.

• For Biologists: It lets them test theories about how animals move without needing to catch rare animals or risk injuring them.
• For Robot Engineers: It lets them test new robot designs in a virtual world before spending money on metal and motors.
• For Everyone: It creates a shared language. If a scientist in Switzerland builds a model, a scientist in the US can download it, tweak it, and run it immediately.

The Bottom Line

FARMS is the ultimate simulation playground. It takes the complex, messy reality of how living things move and turns it into a clean, reusable, and collaborative digital experience. It allows us to ask "What if?" questions—like "What if a robot had a fish's tail?" or "How does a mouse's brain handle a slippery floor?"—and get answers quickly, safely, and accurately.

In short, it's the tool that helps us decode the secret language of movement, one digital step at a time.

Technical Summary

1. Problem Statement

Animal locomotion arises from complex, non-linear interactions between the nervous system, musculoskeletal biomechanics, sensory modalities, and the environment. While experimental data is crucial, it faces significant hurdles:

• Technical Limitations: Measuring forces (e.g., ground reaction forces in small insects) is often infeasible with current technology.
• Ethical Constraints: Certain experiments cannot be conducted on live animals due to ethical concerns.
• Fragmented Modeling Approaches: Existing research often isolates specific aspects of locomotion (e.g., pure robotics, pure biomechanics, or pure neural control).
• Lack of Unified Tools: Current simulation frameworks (e.g., Webots, Gazebo, OpenSim, Isaac Sim) typically specialize in one domain (robotics, biomechanics, or neural networks) but lack a unified, open-source infrastructure to integrate neuromechanical components (muscles, neural circuits, and physics) seamlessly. This forces researchers to duplicate code, hinders reproducibility, and limits the ability to study transitions between different environments (e.g., land-to-water).

2. Methodology: The FARMS Framework

The authors present FARMS (Framework for Animal and Robot Modeling and Simulation), an open-source, modular, Python-based framework designed to unify the workflow for neuromechanical simulations.

Core Architecture

FARMS is structured around four primary stages, connected by a standardized data block (farms_core) that ensures reproducibility and modularity:

1. Data: Manages configuration files (XML/YAML), runtime simulation data, and processed data (HDF5). It allows the state of an experiment to be saved and reloaded at any stage.
2. Modeling: Defines the "Animat" (a unified term for both animals and robots).
   • Body: Uses the Simulation Description Format (SDF) to define rigs (links, joints, inertial properties, and geometries). It supports importing from CAD (robots) or CT/MRI scans (animals).
   • Controller: Supports both model-based control (MPC, inverse dynamics) and neural control (Central Pattern Generators, spiking/non-spiking networks).
   • Actuation: Integrates muscle models (Hill-type, Ekeberg antagonistic pairs) alongside traditional motor actuators.
   • Environment: Supports diverse terrains (flat, rugged) and media (air, water, sand) with specific interaction dynamics.
3. Simulation: Executes the physics loop.
   • Physics Engines: Primarily supports Bullet and MuJoCo. MuJoCo is favored for its implicit damping (crucial for muscle stability) and soft-contact modeling.
   • Fluid Dynamics: Implements drag models and Smoothed-Particle Hydrodynamics (SPH) via the farms_sph package for two-way coupling between fluids and rigid bodies.
   • Loop: The simulation step follows a strict sequence: Sensors $\rightarrow$ External Forces $\rightarrow$ Controller $\rightarrow$ Actuators $\rightarrow$ Physics Step $\rightarrow$ Logging.
4. Analysis: Provides tools for post-processing, visualization, and replay.
   • Blender Integration: A custom add-on (farms_blender) allows for 3D modeling, visualization of simulation replays with overlaid data (forces, contacts), and high-quality rendering for publication.

Key Technical Features

• Modularity: Users can pick specific components (e.g., only the muscle package or only the neural network package) rather than installing the entire suite.
• Extensibility: The framework is designed to wrap existing open-source libraries (e.g., PyTorch, NEST, PySPH) rather than reinventing them.
• Optimization: Includes farms_evolution (based on Pymoo) to tune morphological and control parameters using evolutionary algorithms.

3. Key Contributions

1. Unified Neuromechanical Framework: FARMS is the first open-source framework to seamlessly integrate skeletal structures, complex musculature (Hill-type and Ekeberg models), biological neural networks, and environmental physics (including fluids) in a single workflow.
2. Animal-Robot Parity: It treats biological animals and robots as "Animats," allowing for direct comparison and transfer of control strategies between biological and engineered systems.
3. Fluid-Body Coupling: It provides robust support for simulating aquatic locomotion and transitions (land-water) using drag models and SPH, a feature often missing in standard robotics simulators.
4. Tooling for Complex Models: The farms_blender add-on streamlines the creation of complex musculoskeletal models from CT scans, automating muscle path reconstruction and parameter scaling.
5. Reproducibility: By standardizing data formats (SDF, HDF5) and providing a decoupled pipeline, FARMS significantly lowers the barrier to reproducing complex neuromechanical experiments.

4. Results and Demonstrations

The authors validated FARMS through four distinct case studies:

• Robots (Amphibious Transitions): Simulated Salamandra Robotica II, K-rock, and Pleurobot. The framework successfully replicated transitions from swimming to walking and vice versa using a central pattern generator (CPG) controller. The robots adapted their gaits based on contact sensors and fluid drag forces.
• Biological Salamander: A 3D salamander model (reconstructed from scans) was simulated crossing a river. Unlike the robots (position-controlled), this model used torque-controlled muscles (Ekeberg model). It demonstrated robust walking-to-swimming transitions in a flowing current without explicit terrain mapping, relying on passive muscle compliance and sensory feedback.
• Snake Obstacle Locomotion: A limbless snake model used a simple oscillator network to navigate a grid of pegs. The simulation highlighted how passive body properties and friction interact with obstacles to generate forward motion, a behavior difficult to achieve with rigid control alone.
• Mouse Musculoskeletal Model: A complex, whole-body mouse model with Hill-type musculature and a non-spiking neural circuit was simulated. The study demonstrated the framework's ability to handle high-fidelity biological data, including automated muscle path transfer between skeletal models and closed-loop sensory-driven locomotion on uneven terrain.

5. Significance

• Bridging Disciplines: FARMS acts as a critical bridge between biology, neuroscience, robotics, and computer animation, allowing researchers to test hypotheses about motor control that are impossible to verify experimentally.
• Accelerating Research: By eliminating the need to build simulation infrastructure from scratch, FARMS allows scientists to focus on the science of movement rather than software engineering.
• Scalability: The framework supports everything from simple 2D oscillators to high-fidelity 3D musculoskeletal models with fluid dynamics, making it suitable for a wide range of research scales.
• Open Science: As an open-source, community-driven project, FARMS promotes the sharing of models, data, and control strategies, fostering reproducibility and collaboration in the study of animal and robot locomotion.

In conclusion, FARMS represents a significant step forward in the field of neuromechanics, providing a versatile, extensible, and unified platform to simulate the complex interplay of body, brain, and environment.