Most fluid simulations and robotic control systems rely on a convenient simplification: the Newtonian assumption. This assumes that a fluid's viscosity—its resistance to flow—remains constant regardless of the forces acting upon it. While this works for water or air, it fails for the "messy" fluids that define biology, industry, and the natural environment.
Recent research across medicine, robotics, and geophysics is moving beyond this simplification to model non-Newtonian rheology, where viscosity changes dynamically under stress. By integrating these complex behaviors into computational frameworks, researchers are achieving unprecedented accuracy in everything from artificial heart design to robotic sauce plating.
The Failure of the Newtonian Assumption
In a Newtonian fluid, the relationship between stress and the rate of deformation is linear. However, biological fluids like blood and mucus are non-Newtonian. As Yashaswini Murthy and Ravi N. Banavar demonstrate in A Lagrangian Model to Predict Microscallop Motion in non Newtonian Fluids, these substances often exhibit "shear-thinning" behavior—they become less viscous as they are pushed harder.
This isn't just a theoretical nuance; it dictates physical function. Yuichi Majima and colleagues found in Relationship between mucociliary transport and dry weight of nasal secretion that there is a "Goldilocks" density for mucus transport. If the rheological properties shift too far, the body's natural clearing mechanisms fail. Similarly, Neil W. Bressloff and Prasanth B. Nair argue in Mining data from hemodynamic simulations via Bayesian emulation that ignoring non-Newtonian behavior in blood flow can lead to inaccurate risk assessments for arterial disease.
Differentiable Solvers and Neural Control
The challenge of non-Newtonian fluids is their computational cost. Traditional solvers are often too slow for real-time applications or optimization. To solve this, Yifei Li and their team developed NeuralFluid, a framework that uses differentiable simulation to co-optimize the shape and control of fluidic systems.
By making the fluid solver "differentiable," the system can calculate exactly how a small change in a pump's rhythm or a vessel's shape will affect the flow of a complex substance like blood. This allows for the design of sophisticated devices, such as artificial hearts, that are tailored to the specific rheology of the fluids they move. Similarly, Bing Yan integrated 3D rheological models into oceanographic simulations in Modelling of fluid mud flow based on a three-dimensional hydrodynamic model, significantly improving predictions of how dense sediment flows through estuaries.
Sensing the Unseen via Intuitive Physics
For a robot to interact with the world, it must "sense" these hidden fluid properties. Standard vision systems often struggle with liquids because they look similar but behave differently—honey and oil might appear identical but flow at vastly different rates.
Researchers are bridging this gap through "interactive perception." In Understanding Dynamic Tactile Sensing for Liquid Property Estimation, Hung-Jui Huang and Wenzhen Yuan used high-resolution tactile sensors to "feel" the oscillations of a liquid. By comparing these vibrations against a physics model, the robot can identify a liquid's viscosity with 100% accuracy.
This "intuitive physics" approach is also seen in Kitchen Artist, where a robot tilts a bottle to sense a sauce's viscosity before dispensing it. This allows the robot to adjust its pressure in real-time, compensating for the shear-thinning effects that would otherwise cause a messy splatter.
Bridging the Scale from Micro to Macro
The impact of non-Newtonian modeling spans from the microscopic to the catastrophic. At the micro-scale, Matthew G. Smith and Jack Radford are using machine learning to accelerate microrheology in Machine learning opens a doorway for microrheology with optical tweezers in living systems, allowing scientists to measure the mechanical properties of living cells 1,000 times faster than before.
At the macro-scale, Kaiheng Hu and colleagues demonstrated in Comment on egusphere-2024-884 that standard Newtonian models fail to predict the peak discharge of glacial floods. These disasters involve debris-heavy flows that behave more like liquid mud than water. Only by accounting for variable viscosity can we accurately reconstruct—and eventually predict—these high-altitude disasters.
By treating viscosity as a dynamic variable rather than a static constant, this research direction is providing the tools necessary to navigate, manipulate, and simulate the complex fluids of the real world.