The team, led by Bekir Sami Yilbas, began by creating model hydrophobic surfaces on glass slides using functionalized silica nanoparticles roughly 30 nanometers in size, deposited by dip coating from a solution of silane precursors. Scanning electron microscopy revealed a landscape of nano-sized hills and valleys where trapped air acts as a cushion, shortening the three-phase contact line where liquid, solid and gas meet. The resulting surfaces boasted an average contact angle of 142 degrees with a contact angle hysteresis of only about 2 degrees, meaning droplets roll rather than slide under remarkably small driving forces. A spreading coefficient of minus 106.5 millijoules per square meter confirmed that water refuses to wet the coating, the fundamental prerequisite for self-cleaning behavior.

Dust for the experiments was no laboratory substitute but real environmental dust collected from photovoltaic panels in Dammam, Saudi Arabia, laid down in layers about 150 micrometers thick, matching regional accumulation levels. Energy-dispersive spectroscopy showed the dust contained silicon, calcium, sodium, sulfur, magnesium, potassium, iron and chlorine, with salts and clay-aggregated hematite and gypsum among the constituents. Crucially, when droplet water meets this dust, soluble salt components dissolve: inductively coupled plasma mass spectrometry measured sodium concentrations of about 45,300 parts per billion, with potassium and chlorine at similar levels. This dissolution nudged the surface tension of the fluid from 72 to roughly 72.6 millinewtons per meter and shaved the contact angle from 142 to 141.7 degrees, a small shift with measurable consequences for droplet adhesion.

The central mechanism of dust removal is cloaking, the process by which droplet fluid spreads over and infuses individual dust particles, drawing them into the rolling liquid under capillary action. Using high-speed cameras operating at 300 to 500 frames per second, the researchers measured cloaking velocities averaging around 0.3 millimeters per second, decaying almost exponentially with time. The numbers work out strikingly in favor of cleaning: complete cloaking of particles between 1.2 and 10 micrometers takes about 0.05 seconds, during which a 40-microliter droplet travels only about 34 micrometers, far less than its roughly 2-millimeter contact length on the surface. In other words, the droplet lingers over each patch of dust long enough to swallow it before moving on. An Ohnesorge number of about 0.06 confirmed that viscous shear losses during cloaking are negligible compared with the capillary infusion driving the process.

To peer inside the moving droplet, the team built a three-dimensional computational model coupling the Navier-Stokes equations with heat transfer, evaporation and Marangoni stresses, using a level-set scheme to track the water-air interface. Simulations employed up to 543,827 tetrahedral elements after mesh convergence tests showed velocity and pressure differences below one percent against finer grids, with time steps as small as 10 nanoseconds. The model incorporated a dynamic contact angle, a Furmidge-type retention force capturing contact-angle hysteresis, and temperature-dependent fluid properties. Evaporation was included through a diffusion-convection formulation, though over the short rolling durations simulated the droplet lost negligible volume, so evaporation played only a minor role in the overall dynamics.

The simulations exposed an oscillatory interior world. As the droplet rolls, heat from the inclined surface diffuses upward through the contact area, lowering surface tension locally and generating Marangoni currents that swirl fluid through the droplet interior. Bulk-averaged temperature, velocity and pressure all oscillate in time, with the temperature oscillations tied to slight variations in the contact area caused by droplet wobbling, and the faster velocity oscillations reflecting the fluid’s inertial response. Raising the surface temperature from 25 to 60 degrees Celsius lengthens the three-phase contact line, because warmer liquid spreads more readily, which in turn increases the adhesion force and amplifies the amplitude of wobbling. The location of maximum velocity and pressure inside the droplet shifts continuously with time and temperature, reshaping the wetted footprint as the droplet descends.

Wobbling emerged as a star of the show. The droplet’s center of mass shifts as it rolls, stretching the wetted area into an almost elliptical shape and producing a wavy droplet path across dusty surfaces. High-speed recordings and a mode-classification algorithm applied to the video data revealed that the wobbling resembles the l equals 2 oscillation mode, with frequencies between roughly 40 and 32 hertz for droplets from 10 to 50 microliters, matching theoretical predictions from the Rayleigh-like frequency formula. Predicted normalized apex heights of about 1.42 agreed closely with the experimental value of 1.37. Far from being a nuisance, this wobbling temporarily enlarges the wetted region on each oscillation cycle, actively widening the swath of dust removed from the surface.

Dust, however, exacts a price. Experiments on dusted surfaces showed translational velocities consistently below those on clean surfaces, because cloaking spreads fluid over the particles, enlarging the wetted diameter and adhesion force, increasing interfacial friction, dissolving salts that alter surface tension, and loading the droplet with captured mass. By comparing kinetic energies on dusty and clean surfaces, the team quantified the energy dissipated to dust at about 20 percent. They distilled the relationship into a force ratio scaling with distance along the incline, with a prefactor between 0.7 and 2 and an exponent between 0.19 and 0.23 depending on the tilt angle from 10 to 30 degrees. Meanwhile, the ratio of rotational to translational velocity settled near 0.95, slightly below the value of 1 expected for ideal marble-like rolling, a deficit attributed to dynamic pressure in the surrounding air. Only when the rotational Weber number exceeded about 12 did noticeable slip appear, and even then the slip contribution remained a modest 5 percent of translational velocity.

The cleaning efficiency data carry a practical punch. A 40-microliter droplet removed about 2.1 microliters of dust, while a 20-microliter droplet removed 1.34 microliters, so larger droplets clean more area in absolute terms, yet the dust-to-droplet volume ratio actually falls from about 0.067 to 0.046 as droplets grow from 20 to 60 microliters, because dust removal scales more slowly than droplet volume. The ratio of dust removed by a 10-microliter droplet to that removed by a 40-microliter droplet was about 0.74, underlining how strongly droplet size governs the cleaned footprint. The researchers note that their integrated framework of experiments, three-dimensional simulation and analytical modeling provides a superior physical foundation for self-cleaning in energy harvesting and industrial applications, and they plan to extend the work to multiple-droplet systems, where collective rolling dynamics could transform how solar farms and other dust-exposed infrastructure keep themselves clean.

Subject of Research: Three-dimensional rolling dynamics of water droplets on inclined hydrophobic surfaces and their role in environmental dust removal

Article Title: 3D rolling motion of droplet on inclined hydrophobic surface at different temperatures: Environmental dust removal

Article References: Yilbas, B. S., Hegazy, M. A., Hassan, G., Abubakar, A. A., Al-Qahtani, H., Al-Sharafi, A., & Nouioua, M. (2026). 3D rolling motion of droplet on inclined hydrophobic surface at different temperatures: Environmental dust removal. Results in Engineering, 32, Article 113103. https://doi.org/10.1016/j.rineng.2026.113103

Image Credits: AI Generated

DOI: Not provided

Keywords: hydrophobic surfaces, droplet rolling, self-cleaning, dust removal, Marangoni flow, contact angle hysteresis, cloaking, photovoltaic panels, wobbling oscillations, 3D simulation, surface tension, solar energy

Vision-language navigation, often abbreviated VLN, is the task of instructing an embodied agent, in natural language, to move through an environment, for example, “walk past the kitchen and stop at the door at the end of the hallway.” The agent must ground that instruction in what its cameras see, infer a route, and then physically execute it. Much of the progress in VLN over the past several years has come from models trained in discrete environments, where the world is represented as a graph of navigable nodes connected by edges and the agent simply hops from node to node. That formulation is computationally convenient and has produced impressive instruction-following performance, but it sidesteps a fundamental problem: in the continuous physical world, the space between nodes is filled with furniture, walls, doorframes and other obstacles that a node-hopping agent never has to think about.

The discrepancy between discrete training environments and continuous deployment environments has become a recognized bottleneck. When researchers evaluate discrete-trained VLN models in continuous settings, agents frequently collide with objects, get stuck, or fail to complete their routes, because nothing in their training taught them how to steer around a chair that was never on the navigation graph. Previous efforts to bridge this divide have included waypoint models that convert continuous trajectories into sequences of intermediate targets, and dedicated collision-avoidance frameworks such as Safe-VLN, which trains agents to avoid obstacles in continuous environments. The new paper takes a complementary and distinctly control-theoretic route: rather than retraining the VLN model itself, it wraps the discrete planner inside a layered control system that handles the physical realities of motion.

The architecture is hierarchical in the strict sense. At the top sits a high-level path planner built on VLN models that were trained in a discrete environment. This planner produces an inferred path, essentially a sequence of waypoints that satisfies the language instruction. At the bottom sits a low-level switching control system responsible for actually driving the robot’s wheels along that path. The robot platform considered is a differential-drive robot, the common two-wheeled configuration in which motion is achieved by varying the relative speeds of the left and right wheels. This class of robot is nonholonomic, meaning it cannot slide sideways, which makes precise path following a nontrivial control problem in its own right, independent of any obstacles.

The key innovation lies in how the low-level layer decides which controller to apply at any moment. The system intelligently switches among three classical and modern control strategies: Proportional-Integral-Derivative control, or PID, which is the workhorse of industrial feedback control and works well for steady tracking of a clear path; Proportional control, or P, a lighter-weight variant suited to certain regimes of the trajectory; and Nonlinear Model Predictive Control, or NMPC, an optimization-based technique that, at each control step, solves for the best sequence of future actions subject to constraints on the robot’s dynamics and its environment. NMPC is computationally heavier but far more capable when the robot must re-plan a segment of route that an obstacle has obstructed, because it can look ahead over a prediction horizon and find a feasible detour rather than reacting myopically.

Switching between controllers is not a trivial engineering choice. Control theorists have long studied switched systems, and a classical result by Hespanha and Morse established that stability of a switched system depends on how frequently and how abruptly the switching occurs; switch too erratically and even individually stable controllers can produce unstable overall behavior. The paper’s switching logic is designed with this in mind, selecting the active controller based on the robot’s situation, whether it is tracking an unobstructed path segment or confronting a detected obstacle, so that transitions between modes remain well-behaved. Depth sensing plays a central role here: by perceiving the distance to obstacles in the robot’s field of view, the system knows when a planned segment is blocked and when it is safe to hand control back to the smooth path-following controllers.

When the depth sensors reveal a static obstacle blocking the inferred route, the system does not simply stop or nudge around it with ad-hoc rules. Instead, the NMPC layer dynamically re-plans the obstructed segment of the route, generating a local detour that respects the robot’s kinematic constraints, and the robot then rejoins the original path beyond the obstruction. This division of labor means the expensive VLN model does not need to be retrained or re-queried every time a sofa sits where the planner expected clear floor. The intelligence about the world’s clutter lives in the control layer, where it can be handled with fast, well-understood feedback techniques running at control-loop rates rather than at the slower cadence of a large neural network.

The authors validated the approach in a realistic three-dimensional simulator, testing the full pipeline from language instruction to VLN-based path inference to switched-control execution with obstacle detours. The headline result is a substantial reduction in the number of collisions per episode, dropping from tens of collisions in the baseline setting to fewer than one on average. That order-of-magnitude improvement matters because collisions are not merely cosmetic failures; each one can leave an agent stuck, invalidate the rest of its plan, or, on physical hardware, damage the robot or its surroundings. The simulation code has been released publicly on GitHub, allowing other researchers to reproduce and build on the results.

The broader significance of the work is that it offers a viable recipe for bridging the gap between discrete high-level navigation plans and continuous robot control, without discarding the large investment the community has made in discrete-environment VLN models. Because the collision-handling intelligence is modular, improvements in VLN models, whether from better multimodal pre-training, larger language backbones, or new training objectives, can be dropped into the high-level layer while the control layer continues to guarantee safe execution. The authors frame the contribution as enhancing the robustness of VLN agents in cluttered or obstructed working spaces, a property that will be essential for assistive robots, delivery robots and any embodied agent expected to follow instructions in environments that were not engineered to be empty. As embodied AI moves from benchmark graphs toward homes, hospitals and warehouses, architectures like this one, which marry learned semantic navigation with rigorously engineered control, are likely to define the path from simulation success to real-world reliability.

Subject of Research: Bridging discrete vision-language navigation planning and continuous robot control using switched controllers and depth sensing to reduce collisions

Article Title: Discrete to continuous VLN adaptation with collision mitigation through depth sensing

Article References: Khan, J., Salahuddin, H., Aafaq, N., & Ali, Q. (2026). Discrete to continuous VLN adaptation with collision mitigation through depth sensing. International Journal of Intelligent Robotics and Applications. https://doi.org/10.1007/s41315-026-00592-1

Image Credits: AI Generated

DOI: 10.1007/s41315-026-00592-1

Keywords: vision-language navigation, VLN, robotics, differential-drive robot, collision avoidance, depth sensing, model predictive control, PID control, switched systems, embodied AI, path planning, 3D simulation