The team’s starting point was a full six-degree-of-freedom missile model, the standard simulation framework for capturing how an air-to-ground weapon translates and rotates in three dimensions under the combined influence of thrust, gravity, drag, and aerodynamic moments. Within this virtual environment, the researchers kept the guidance architecture conventional where it worked well: yaw was handled through waypoint navigation, and pitch followed Proportional Navigation Guidance, the time-honored law that steers an interceptor by closing the line-of-sight rotation rate toward zero. The roll channel, however, was the battleground. A baseline proportional-integral-derivative controller, the workhorse of flight control for decades, provided the reference performance against which every artificial intelligence approach would be measured.

Roll control matters more than casual observers might assume. For an axisymmetric missile, roll couples into pitch and yaw dynamics, meaning that an uncontrolled roll can smear the seeker’s view of the target, distort the effectiveness of canard or fin deflections, and induce structural loading that a slender airframe was never designed to absorb. Prior research has documented how elastic deformation, wrap-around fins, and spin-deformation coupling all conspire to make the roll channel notoriously difficult, particularly for small air-to-surface missiles where actuator authority is limited and disturbances arrive fast. The Turkish-Indonesian team set out to see whether learned controllers could integrate sensor alignment with structural integrity constraints better than a tuned PID loop, especially when the missile is forced into aggressive terminal maneuvers.

The researchers trained and compared four AI strategies. Linear regression served as the simplest baseline, learning a direct mapping from flight states to roll control inputs. A neural network built on long short-term memory units brought the ability to retain information over time, a natural fit for a dynamical system whose history matters. Deep Deterministic Policy Gradient, a reinforcement learning algorithm that continuously refines a control policy through interaction with the environment, added the capacity to discover control strategies that no human engineer would hand-code. Finally, the team built an ensemble combining all three, letting the strengths of regression, recurrent memory, and reinforcement learning compensate for one another’s weaknesses. Crucially, all models were trained on data generated by the PID controller itself, meaning the AI systems learned from the accumulated experience of the classical controller before surpassing it.

Testing took place under two families of scenarios: straight flight toward the target, and aggressive yaw maneuvers involving 90-degree and 180-degree turns during the terminal phase. Performance was scored on two fronts. The first was roll angle deviation, a direct measure of how well each controller kept the missile’s roll attitude near its commanded value. The second was terminal miss distance, the ultimate metric in guided munitions, capturing how close the weapon came to its intended impact point. A controller that holds roll beautifully but misses the target by meters is useless; one that hits within centimeters while keeping the airframe stable is the goal.

The ensemble method emerged as the clear winner. In straight flight it delivered near-minimal roll error, essentially matching the best achievable stabilization. During the punishing 90-degree and 180-degree yaw maneuvers, where rapid changes in lateral acceleration ripple through the roll dynamics, the ensemble maintained superior stability while the simpler models struggled. Most strikingly, it achieved the lowest miss distance recorded in the study, 0.14 meters, in the most demanding scenarios. For context, a miss distance measured in tens of centimeters rather than meters represents the difference between a weapon that reliably defeats a target and one that may require multiple engagements.

Deep Deterministic Policy Gradient on its own also proved robust, posting consistently low miss distances and handling the nonlinearities and disturbances that plague the terminal phase with notable composure. This is consistent with a broader trend in aerospace research, where reinforcement learning controllers have shown an appetite for exactly the kinds of high-dimensional, nonlinear problems that frustrate classical gain-scheduled autopilots. The reinforcement learning agent, in effect, learns through trial and error within the simulation which control actions keep the roll channel quiet under conditions no single linear model can describe, and it can do so without an engineer manually re-tuning gains for every flight regime.

The weaker performers were equally instructive. Linear regression and the LSTM network exhibited noticeably higher roll errors and miss distances, indicating limited adaptability to rapid changes in flight conditions. The lesson is not that machine learning fails, but that the choice of architecture matters enormously. A static regression model can interpolate within the data it has seen but has little capacity to respond gracefully when the missile’s dynamics depart from its training distribution, as they inevitably do during a violent turn. Even a recurrent network with memory, trained purely on PID-generated trajectories, appears to inherit some of the classical controller’s limitations. The ensemble’s success suggests that diversity of learning paradigms, rather than any single clever algorithm, is what buys robustness at the edge of the flight envelope.

The implications extend beyond munitions. The same roll-stabilization problem appears in reusable launch vehicles, hypersonic gliders, and small autonomous aircraft, and the study’s authors sit squarely within a research community, spanning institutions from Marmara University to Indonesia’s National Research and Innovation Agency, that has been systematically probing where AI control can replace or augment classical loops. Their work was supported by Turkey’s TÜBİTAK 2209-A university research program and Indonesia’s Rumah Program, and it follows earlier contributions from the same group on multi-phase missile guidance and reinforcement learning for landing guidance. The pattern across this literature is consistent: learned and hybrid controllers tend to shine precisely where dynamics are fastest and most nonlinear, which is to say where classical control is weakest.

Caveats remain, and the authors are candid about them. Everything reported here is simulation. A six-degree-of-freedom model, however sophisticated, cannot fully capture sensor noise, actuator saturation and lag, structural flexing, or the computational constraints of real flight hardware. The researchers explicitly state that real-world hardware validation remains essential to confirm the simulation outcomes and to support any operational deployment, a caution that echoes throughout the field of AI-augmented flight control. Verification and certification of learned controllers is an unsolved regulatory problem, and no air force will fly a neural network it cannot formally verify. Still, the trajectory of the evidence is hard to ignore. When an ensemble of regression, memory networks, and reinforcement learning can hold a simulated missile steady through a 180-degree terminal turn and guide it to within 14 centimeters of its aim point, the classical autopilot has been put on notice. The next step is proving it in the air.

Subject of Research: AI-based roll stabilization control for air-to-ground missiles during dynamic terminal flight

Article Title: AI-Augmented Roll Stabilization for Missiles Under Dynamic Terminal Flight Conditions

Article References: Abdurrauf, A., Nugroho, L., & Ülkü, E. E. (2026). AI-Augmented Roll Stabilization for Missiles Under Dynamic Terminal Flight Conditions. International Journal of Aeronautical and Space Sciences. https://doi.org/10.1007/s42405-026-01233-4

Image Credits: AI Generated

DOI: 10.1007/s42405-026-01233-4

Keywords: missile guidance, roll stabilization, artificial intelligence, reinforcement learning, DDPG, LSTM neural networks, PID control, six-degree-of-freedom simulation, terminal flight, miss distance, flight control systems, ensemble learning

The central challenge the researchers tackled is a fundamental tension in flight control. On one hand, engineers want to guarantee that the aircraft’s tracking errors stay within strict bounds, a property known as prescribed performance, so that the jet never strays outside a safe corridor during its descent. On the other hand, the aircraft’s control surfaces and engine thrust are physically limited. When a controller demands more deflection or thrust than the actuators can deliver, a condition called actuator saturation, the aircraft temporarily stops obeying commands and the error can breach the very boundaries the performance function was designed to enforce. Once those boundaries are violated, many prescribed performance controllers fail catastrophically, because their mathematics assumes the error always remains inside a constrained domain.

The Harbin team’s solution is elegantly pragmatic: rather than pretending saturation will never happen, they built a controller that bends without breaking. At the heart of the method lies a predefined-time prescribed performance function, a mathematical curve that dictates how quickly and how tightly the landing error must shrink. Unlike conventional exponential convergence guarantees, which depend on initial conditions, a predefined-time guarantee ensures the error converges within a time chosen by the designer, no matter where the aircraft starts. This is a significant advantage in carrier operations, where the approach window is short and unforgiving and where an autopilot that converges ‘eventually’ is not good enough.

To work with these constraints mathematically, the researchers employed an error transformation technique that converts the constrained error domain into an equivalent unconstrained one. In essence, the transformation warps the coordinate system so that as long as the transformed error remains bounded, the physical landing error is guaranteed to stay inside its prescribed funnel. This allows standard control design tools to be applied to a problem that would otherwise involve hard, nonlinear boundaries. The transformed error then feeds into a nonsingular fast terminal integral sliding mode controller, a robust control strategy renowned for its ability to reject external disturbances. The ‘terminal’ component drives errors to zero in finite time rather than merely asymptotically, while the ‘nonsingular’ design avoids the mathematical pitfalls, such as division by zero, that plague naive terminal sliding mode formulations.

The truly novel ingredient, however, is the flexible adjustment mechanism embedded within the performance function itself. When the controller detects that actuators are saturating, meaning the aircraft cannot fully execute the commanded deflections, the mechanism adaptively relaxes the local performance boundaries, giving the aircraft room to recover rather than demanding the impossible. Once the saturation episode passes, the boundaries tighten again, and the landing proceeds with its original accuracy targets intact. This adaptive flexibility prevents the boundary violations that would otherwise cause the error transformation to blow up and the controller to fail. Complementing this mechanism, the team designed an auxiliary system that maintains stability during saturation events, a well-known strategy in the control literature that the authors integrated carefully with their performance framework.

Disturbance rejection receives equally thorough treatment. The turbulent airwake behind a carrier’s island superstructure is modeled in the study as a set of external disturbances acting on the aircraft’s translational and rotational dynamics, with components that vary sinusoidally over time to mimic the chaotic flow field. To counter these, the researchers developed a predefined-time disturbance observer, an algorithm that estimates the unknown disturbance signals and feeds the estimates directly into the control law as compensation. Because the observer converges in predefined time, the compensation kicks in quickly enough to matter during the brief final approach. The team also introduced a predefined-time reference model to generate the command signals that the inner control loops track. This choice eliminates a chronic computational problem in conventional backstepping control designs, where repeated differentiation of command signals leads to ‘explosion of complexity,’ an exponential growth in the analytic terms that must be computed at each step.

The aircraft model used in the study is a full six-degree-of-freedom representation of a carrier-based aircraft, complete with the coupling between roll, pitch, and yaw dynamics that arises from the aircraft’s inertia properties, aerodynamic coefficients for lift, drag, and side force, and the contributions of the elevator, aileron, rudder, and direct lift flap. The guidance problem is framed as following a moving deck landing path, with the controller tasked with tracking the ship’s motion while respecting output constraints on altitude and attitude errors. This comprehensive modeling matters, because simplified models often hide exactly the coupling effects that make real carrier approaches so treacherous, particularly the way lateral gusts induce rolling moments that must be countered within fractions of a second.

To benchmark their approach, the authors compared FPTPPC against three competing strategies in simulation: dynamic inverse control, a classical technique that cancels the aircraft’s nonlinearities through model inversion; predefined-time control without prescribed performance; and standard prescribed performance control without the flexible saturation mechanism. The results showed that the proposed method reduced path following errors and improved overall landing accuracy relative to all three baselines. The simulations also demonstrated that the flexible adjustment mechanism kept the controller stable and accurate through saturation events that would have destabilized a rigid prescribed performance design. While the study is computational rather than flight-tested, the consistency of the improvements across the comparison set lends weight to the practical promise of the approach.

The implications extend beyond naval aviation. The core ideas, predefined-time convergence guarantees, flexible performance boundaries that tolerate actuator limits, and disturbance observers that compensate within a designer-chosen horizon, are directly transferable to other safety-critical control problems, including spacecraft attitude control, unmanned aerial vehicle operations in gusty conditions, and robotic systems with limited actuator authority. Related work in the same journal and elsewhere has begun applying predefined-time and anti-saturation techniques to fault-tolerant flight control, suggesting a broader research movement toward controllers that are honest about physical limits rather than assuming them away. For the moment, the Harbin team’s contribution stands as a careful piece of control engineering: a mathematically rigorous answer to the question of what an autopilot should do when the ideal command and the possible command diverge, at the worst possible moment, a few hundred feet above a moving flight deck.

Subject of Research: Automatic carrier landing control using flexible predefined-time prescribed performance control under output constraints, actuator saturation, and airwake disturbances

Article Title: Flexible Predefined-Time Prescribed Performance Control for Automatic Carrier Landing with Output Constraints and Actuator Saturation

Article References: Wang, Z., Zhu, Q., Kong, L., & Wang, L. (2026). Flexible Predefined-Time Prescribed Performance Control for Automatic Carrier Landing with Output Constraints and Actuator Saturation. International Journal of Aeronautical and Space Sciences. https://doi.org/10.1007/s42405-026-01239-y

Image Credits: AI Generated

DOI: 10.1007/s42405-026-01239-y

Keywords: automatic carrier landing, prescribed performance control, predefined-time control, actuator saturation, sliding mode control, airwake disturbance, flight control systems, disturbance observer, carrier-based aircraft, nonlinear control, landing accuracy, Harbin Engineering University

The research, published in the International Journal of Aeronautical and Space Sciences by Qilong Wu of Huainan Normal University, Qidan Zhu of Harbin Engineering University, and their colleagues, including Lunpan Wei, Lipeng Wang, and Wenqiang Jiang of the China Ship Development and Design Center, tackles a problem that has long frustrated engineers: how to guarantee that a landing aircraft stays on its glide path within a strictly bounded error, and that it does so within a time window chosen in advance by the designer, regardless of where the aircraft starts its approach. The work was supported by the National Natural Science Foundation of China and other national and provincial funding bodies.

At the heart of the new scheme is what the authors call a predefined-time prescribed performance function, or PTPPF. Prescribed performance control is a technique that constrains the tracking error, the difference between where the aircraft should be and where it actually is, to remain inside a shrinking envelope. Think of it as a funnel around the ideal glide slope: the aircraft’s deviation is mathematically forbidden from escaping the funnel walls, and the funnel narrows as the aircraft nears the deck. What makes the new formulation distinctive is the predefined-time element. In many conventional designs, the convergence of the error depends on the initial conditions, meaning a jet that starts its approach far from the ideal path may take an unpredictable amount of time to settle. The PTPPF removes that dependency entirely, guaranteeing that the error converges within a time set by the control engineer before the mission even begins.

Guaranteeing a bounded error is only half the battle, however. The second pillar of the approach is a predefined-time sliding mode controller, or PTSMC. Sliding mode control is a well-established robust technique that drives a system’s state onto a chosen sliding surface and then forces it to slide along that surface toward the target, effectively ignoring matched uncertainties and external disturbances along the way. Its weakness has traditionally been the chattering phenomenon, in which the control signal oscillates rapidly, and the fact that classical finite-time versions offer convergence speeds that vary with initial conditions. The predefined-time variant designed by the team enhances robustness against the nonlinear uncertainties and external disturbances that plague carrier approaches, while ensuring that the sliding dynamics complete their work within the designer’s chosen deadline.

The third and perhaps most ingenious component is a prescribed-time extended state observer, or PTESO. Extended state observers are devices that treat all the unknown disturbances acting on a system, including unmodeled dynamics, wind gusts, and even actuator faults, as an additional augmented state to be estimated alongside the system’s physical states. Once estimated, these lumped disturbances can be cancelled out in real time within the control loop, a strategy known as active disturbance rejection. The prescribed-time version developed here goes further: it promises that its estimates of the combined disturbances and actuator faults become accurate within a prescribed time, allowing the controller to compensate for a failing elevator or a sudden shift in the ship’s turbulent wake almost as soon as it occurs, rather than gradually over the course of the approach.

To prove that this elaborate architecture is safe rather than merely plausible, the authors carried out a Lyapunov-based stability analysis. Lyapunov methods are the gold standard of nonlinear control theory: the designer constructs an energy-like function of the closed-loop system and demonstrates mathematically that this function decreases along the system’s trajectories, which implies that all signals remain bounded and that the tracking error behaves as intended. In this case, the analysis establishes two critical guarantees simultaneously: every closed-loop signal stays bounded, and the tracking error converges within the predefined time, no matter the initial conditions. That combination of boundedness and deadline-driven convergence is what separates this work from earlier prescribed performance designs, which typically offered only asymptotic or condition-dependent convergence.

The true test of any carrier landing control law comes in simulation that faithfully reproduces the hostile environment of a moving flight deck. The team’s carrier landing experiments incorporated three of the most punishing factors at once: the carrier air-wake, the turbulent and horizontally offset airflow generated by the ship’s island superstructure as it washes over the approaching aircraft; the deck motion, meaning the heave, pitch, and roll of the landing point as the ship responds to waves; and actuator faults, including partial loss of control effectiveness. These are precisely the coupled hazards that have caused past approaches to degrade, because a controller tuned for calm conditions can be overwhelmed when disturbances and hardware failures arrive together.

The results, according to the authors, show that the proposed method significantly improves transient performance, disturbance rejection, and landing accuracy compared with existing techniques. Transient performance matters enormously in this context: the moments right after the aircraft enters the glide path, or right after a fault occurs, are when deviations are largest and most dangerous. By confining the error inside the prescribed performance funnel from the very start and compensating disturbances through the observer in real time, the scheme keeps those dangerous excursions small and short-lived. The practical implication is a tighter, more predictable touchdown point on the deck, which translates directly into reduced structural stress on the airframe and lower risk of missing the arresting wires.

The significance of the predefined-time formulation becomes clearer when one considers the communication environment referenced in the study’s title. Carrier-based aircraft operate in data links with the ship, and guidance updates may arrive late or intermittently. A control law whose convergence time balloons when conditions are unfavorable offers no firm promise to the pilot or the mission planner. By contrast, a scheme in which the error is guaranteed to converge within a time fixed at design stage, independent of initial conditions, allows the entire approach to be choreographed against a known clock, which is essential when the landing window is dictated by the ship’s motion and the recovery schedule of the air wing.

The work also fits into a broader and rapidly accelerating research trend. Fixed-time and predefined-time control methods have recently swept through the control theory literature, with applications demonstrated for multiagent consensus, microgrid frequency regulation, hypersonic vehicles, spacecraft attitude stabilization, and quadrotor drones, including vehicles suffering actuator failures. Meanwhile, extended state observers have been combined with sliding mode and super-twisting techniques for carrier landing in a series of recent studies, and fault-tolerant control for carrier-based aircraft has evolved from fault reconstruction schemes to adaptive neural approaches. The new paper synthesizes these strands, wrapping a predefined-time prescribed performance function, a predefined-time sliding mode controller, and a prescribed-time extended state observer into a single fault-tolerant architecture with formal stability proofs.

For now, the results rest on simulation experiments rather than flight tests, and the authors note that the underlying code is not publicly available due to project confidentiality and institutional restrictions, though it can be requested from the corresponding author for academic purposes. The data supporting the findings are likewise available upon reasonable request with laboratory approval. Those caveats are standard for research at this stage, and the gap between simulated and real carrier operations remains a formidable one, involving sensor noise, actuator saturation limits, and the sheer unpredictability of the maritime environment. Nevertheless, the mathematical guarantees established in the paper, bounded closed-loop signals and deadline-driven error convergence under simultaneous air-wake disturbance, deck motion, and actuator faults, represent a meaningful step toward the long-sought goal of fully autonomous, fault-resilient carrier recoveries. If future flight validation confirms the simulation results, the technology could one day allow unmanned carrier-based aircraft, and perhaps crewed ones, to come aboard in conditions that today demand the finest human piloting skills in existence.

Subject of Research: Fault-tolerant predefined-time control for automatic aircraft carrier landing under air-wake disturbances and actuator faults

Article Title: Robust Predefined-Time Fault-Tolerant Prescribed Performance Control for Automatic Carrier Landing in Communication Environments Under Air-Wake Disturbances and Actuator Faults

Article References: Wu, Q., Zhu, Q., Wei, L., Wang, L., & Jiang, W. (2026). Robust Predefined-Time Fault-Tolerant Prescribed Performance Control for Automatic Carrier Landing in Communication Environments Under Air-Wake Disturbances and Actuator Faults. International Journal of Aeronautical and Space Sciences. https://doi.org/10.1007/s42405-026-01266-9

Image Credits: AI Generated

DOI: 10.1007/s42405-026-01266-9

Keywords: automatic carrier landing, fault-tolerant control, predefined-time control, sliding mode control, prescribed performance, extended state observer, air-wake disturbance, actuator faults, Lyapunov stability, carrier-based aircraft, flight control systems, disturbance rejection