Chapter Four · failure evidence
What Adaptive Control got wrong, from 35 dissertations
The records document various practical and theoretical failure modes encountered when developing and implementing adaptive control methods. Common difficulties include mathematical instabilities under unmodeled dynamics or high adaptation rates, degradation from discontinuous switching laws, and underperformance relative to simpler fixed-gain baselines. These records come from PhD theses at 18 institutions, 2021 to 2026. Each links to its thesis. They were extracted by language models reading the full text, so treat each as a lead to read, not a verdict.
Adaptive schemes become unstable under unmodeled dynamics, structural changes, or parameter variations
Standard and linearized model reference adaptive controllers failed or were rejected because they could not compensate for unmatched uncertainties, structural reconfigurations, or unmodeled nonlinear coupling. Time-varying parameters and multi-threaded architectures also injected non-vanishing indefinite terms that violated Lyapunov monotonicity and broke stability.
Tried and failed
multi-thread attracting manifold adaptive control applied to aerospace dynamical systems. Outcome: unstable. Reason: indirect adaptation violated strict Lyapunov monotonicity across simultaneous threads, producing sign-indefinite derivatives
Multi-threaded attracting manifold adaptive control for aerospace systems · UT Austin
Tried and failed
standard model reference adaptive control applied to systems with unmatched parameter uncertainties. Outcome: unstable. Reason: standard MRAC cannot stabilize systems under unmatched uncertainties from payload variations
Cooperative control of multi-uavs under communication constraints. · Cranfield
Considered and rejected
Considered and rejected: Rejected traditional PD and linearized model-reference adaptive controllers because they cannot compensate for unmodeled nonlinear coupling dynamics or terrain variations without destabilizing.
Achieving Near-Natural Locomotion in Transfemoral Amputees - A Control Theoretic Approach · unevada
Tried and failed
classical adaptive control schemes applied to persistently time-varying parameter systems. Outcome: unstable. Reason: parameter time derivatives inject non-vanishing indefinite terms, breaking passivity and L2 stability properties
Adaptive control for time-varying systems: congelation and interconnection · Imperial
Tried and failed
single-model reference adaptive control applied to switched dynamic systems. Outcome: unstable. Reason: parameters tuned for one configuration caused tracking divergence after structural reconfiguration
Routing and Control of Unmanned Aerial Vehicles for Performing Contact-Based Tasks · Virginia Tech
Tried and failed
L1 adaptive control applied to quadrotor tracking under large parametric uncertainty. Outcome: unstable. Reason: parametric error grew too large in out-of-distribution conditions, causing instability and tracking degradation
ONLINE CONTROLLER ADAPTATION WITH META-LEARNED MODELS · Penn
Adaptive controllers underperform simpler baselines or adapt too slowly for dynamic environments
Multiple adaptive controllers delivered inferior tracking accuracy, higher latency, or worse disturbance filtering compared to standard PID or feedback linearization baselines. Other adaptive formulations were rejected or degraded because they could not adapt quickly enough to non-stationary environments or lacked critical state information.
Lost to a baseline
Under noise dissonance sensor faults, standard PID controllers filtered disturbances and achieved lower initial RMSE faster than adaptive MRAC architectures.
Enhanced Model Reference Adaptive Controllers for Fault-Tolerant Controls in Industrial Applications · DeustoTeka
Tried and failed
adaptive cautious control with tight parameter bounds applied to run-to-run lithography process control. Outcome: worse than baseline. Reason: tight control bounds reduce cautious controller improvement rate as process gain uncertainty grows
Advanced process control in manufacturing process with high dimensional measurements · Georgia Tech
Lost to a baseline
In physical experiments with attached slung payloads, L1-MPC-Int performed relatively worse than KNODE-based adaptive variants (L1-MPC x-y RMSE was 22.0% to 24.6% higher) because its state predictor lacked position states and could not compensate for motion capture position uncertainties.
Learning-based Model Predictive Control for Aerial Vehicles · Penn
Lost to a baseline
Steer-by-Wire actuator detection latency was 44.2 ms for the proposed adaptive approach vs. 27.2 ms for the Kalman check with constant coding vector
REAL-TIME ERROR DETECTION AND CORRECTION FOR ROBUST OPERATION OF AUTONOMOUS SYSTEMS USING ENCODED STATE CHECKS · Georgia Tech
Lost to a baseline
Feedback linearization controller achieved lower position RMS tracking error (0.287) than adaptive sliding mode control (0.327) on the nominal AFF trajectory when mass fraction was zero / perfectly known.
Online Information-Aware Motion Planning with Model Improvement for Uncertain Mobile Robotics · MIT
Considered and rejected
Considered and rejected: Rejected filtered-x LMS (FXLMS) adaptive noise cancellation because it could not adapt rapidly enough to non-stationary dynamic clinic noise.
Improving Respiratory Sound Monitoring and Analysis through Noise Control, Sensor Design, and Real-World Considerations · JScholarship
Adaptive controllers are rejected due to drift risks, certification barriers, and real-time complexity
Designers rejected continuous parameter adaptation without triggers to prevent parameter drift, bursting, and random walks in unexcited states. In addition, adaptive controllers faced rejection due to flight certification hurdles regarding predictability as well as excessive computational load and complexity for real-time operation.
Considered and rejected
Considered and rejected: Adaptive PID control due to certification issues (FAR/CS-23/25, DO-178C) regarding unpredictability, parameter variations, and synchronization issues in redundant architectures
Compact Electromechanical Actuators for Urban Air Mobility: Development of a Framework for Design, Digital Twin and PHM Integration in eVTOL Aircrafts · IRIS - POLITO - prod
Considered and rejected
Considered and rejected: Rejected continuous parameter learning/adaptive control without triggering due to risk of parameter drift/random walks in unexcited states and bursting.
Event-triggered Learning · Publikationssystem UB Tuebingen
Considered and rejected
Considered and rejected: Rejected explicit parameter adaptive control due to the inability to guarantee stability if online parameter estimates are imperfect and the requirement of designing complex online identifiers.
Aircraft flight control system design by model reference adaptive control · Iowa State
Considered and rejected
Considered and rejected: Adaptive controllers for payload compensation were rejected because they 'may suffer from over-fitting, slow convergence, excessive computational load for real-time implementations'.
Novel Technology Perspectives for Urban Air Mobility Applications · IRIS - POLITO - prod
Considered and rejected
Considered and rejected: Rejected adaptive oscillators, continuous model-based methods, and volitional control in favor of speed-scaled impedance finite-state machines to avoid real-time nonlinear dynamics complexity and inconsistent volitional inputs.
Enabling Personalized Prosthetic Control Using Self-Learning and Bayesian Optimization · Georgia Tech
Excessively high adaptation rates and gains trigger instability or amplify noise
Selecting high adaptation rates or aggressive droop gains to accelerate convergence caused dynamic instability and exceeded inertia and trajectory bounds. These aggressive gains amplified high-frequency sensor noise and unmodeled dynamics, which in some cases produced secondary frequency dips.
Tried and failed
adaptive attitude control with constant gains applied to spacecraft attitude tracking. Outcome: unstable. Reason: adaptation gain exceeds inertia and trajectory bounds
Onboard control, tracking and navigation for autonomous systems · UT Austin
Tried and failed
adaptive feedforward control with high learning rates applied to robot motion control with sensory feedback. Outcome: unstable. Reason: Excessively high adaptation rate in the feedforward loop caused dynamic control instability across trials.
Bio-inspired robotics: efference copies and adaptive feedforward control · Imperial
Tried and failed
high adaptation learning rates for faster convergence applied to adaptive control of dynamical systems. Outcome: unstable. Reason: amplified sensor noise and unmodeled dynamics or disturbances
Multi-threaded attracting manifold adaptive control for aerospace systems · UT Austin
Tried and failed
large fixed adaptive droop gain applied to wind turbine frequency support. Outcome: worse than baseline. Reason: steep electrical power drop as rotor speed declined caused secondary frequency dip
Discontinuous adaptation laws and non-smooth mechanisms induce chattering and oscillations
Non-smooth saturation projections and signum-based switching induced high-frequency parameter bouncing and control torque oscillations on physical hardware. Fixed-step integration also produced severe numerical chattering when dividing by near-zero regressor norms during parameter estimation.
Tried and failed
non-smooth saturation projection in adaptive control applied to parameter estimation in dynamical systems. Outcome: unstable. Reason: discontinuous projection caused parameter bouncing and closed-loop oscillations exceeding actuator bandwidth limits
An adaptive control framework with applications to intelligent and human-centric vehicular automation · UT Austin
Tried and failed
Concurrent learning model reference adaptive control applied to dynamic structural parameter estimation. Outcome: unstable. Reason: Transient state transitions between equilibrium states caused severe parameter chattering and overshoots
Integrated smart sensor networks with adaptive real-time modeling capabilities · Iowa State
Tried and failed
finite-time adaptive parameter estimation applied to online system identification. Outcome: unstable. Reason: fixed-step integration caused severe chattering when dividing by near-zero regressor norms
Aircraft System Identification Approach for Control Surface Fault Diagnosis · Virginia Tech
Tried and failed
signum-based adaptive disturbance upper bound estimation applied to nonlinear attitude tracking control. Outcome: unstable. Reason: Discontinuous signum switching induced high-frequency marginally stable control torque oscillations on physical hardware.
Adaptive Controller Development and Evaluation for a 6DOF Controllable Multirotor · Virginia Tech
Actuator saturation limits, frequency thresholds, and simulation mismatches cause hardware failures
Adaptive controllers drove actuators into saturation limits during transient tracking regulation or destabilized when operating frequencies crossed critical limits. Furthermore, gains tuned purely in simulation failed to generalize to hardware because of unmodeled actuator dynamics and aerodynamic interactions.
Tried and failed
Model reference adaptive control with prescribed performance control applied to constrained nonlinear tracking control. Outcome: unstable. Reason: Transient tracking error regulation caused control inputs to exceed physical actuator saturation limits.
Routing and Control of Unmanned Aerial Vehicles for Performing Contact-Based Tasks · Virginia Tech
Tried and failed
simulation-based tuning of adaptive controller gains applied to multi-rotor aerial vehicle flight control. Outcome: did not generalise. Reason: unmodeled actuator dynamics and aerodynamic downwash interactions caused mismatch between simulation and hardware
Adaptive Controller Development and Evaluation for a 6DOF Controllable Multirotor · Virginia Tech
Tried and failed
linear adaptive feedforward control applied to quadrotor position tracking under asymmetric disturbance. Reason: linear feedforward components cannot generate the harmonic frequencies absent from reference inputs to cancel asymmetric disturbances
Bio-inspired robotics: efference copies and adaptive feedforward control · Imperial
Tried and failed
adaptive neural network bilateral impedance control applied to teleoperated upper-limb robotic exoskeletons. Outcome: unstable. Reason: increasing the motion frequency beyond a critical threshold induced system instability
Left open by the authors
Problems the authors named and did not get to.
Left open
Extend composite adaptive backstepping control to handle time-varying parameters, unstructured uncertainties, and unmatched uncertainties in flight dynamics simulation. Blocker: None
Nonlinear flight control with reduced model dependency · Cranfield
Left open
Implement adaptive control for chained Stewart platform models to maintain robustness against unknown payload mass and actuator failures in simulation. Blocker: None
Methods for Kinematic Analysis and Optimization of Overactuated Serial and Parallel Structures · Virginia Tech
Left open
Implement cascaded adaptive visual predictive control on a free-flying multicopter to reject exogenous disturbances like wind gusts. Blocker: Requires a physical free-flying multicopter platform and test environment with controlled disturbances (e.g., wind generation).
Predictive visual servoing; uncertainty analysis and probabilistic robust frameworks · oURspace
Left open
Design an adaptive control scheme to automatically tune controller delay and gain along a stable path during quasi-steady state equilibrium shifts. Blocker: None
Small-Signal Stability Techniques for Power System Modal Analysis, Control, and Numerical Integration · Research Repository UCD
Left open
Develop an adaptive event-triggered control formulation using Control Barrier Functions for safety-critical systems. Blocker: The task is described only as a high-level research direction without specific formulations or targets
Optimal control and learning for safety-critical autonomous systems · OpenBU
Left open
Implement adaptive PID controller parameter scheduling based on proximity to the target goal for the autonomous blimp robot. Blocker: Requires the physical blimp robotic platform or custom robotic hardware setup for tuning and validation
Feedback Controller Design For Low-Power Autonomous Blimp Robot · MIT
Left open
Develop adaptive control policies for legged robots that self-refine over time to compensate for physical hardware wear without manual recalibration. Blocker: None
A Model-Based Planning and Control Framework for Parkour-Style Legged Locomotion · MIT
Left open
Incorporate performance dose-response models into an intelligent feedback control system using model predictive and adaptive control updating parameters per time step. Blocker: None
Optimization and Adaptation of Athlete Training Load With Evolutionary Computation · Research Repository UCD
Left open
Extend the thesis's adaptive control framework to include the gripper link motor on the robotic arm in simulation. Blocker: None
Adaptive control of a DDMR with a Robotic Arm · Virginia Tech
Left open
Develop and implement adaptive Smith-predictor-based feedback controllers for improved setpoint regulation in rolling-diaphragm hydrostatic transmission actuators. Blocker: Requires the physical hydrostatic transmission robotic hardware or proprietary experimental setup for validation
Rolling-diaphragm hydrostatic transmission for the remote actuation of high-performance robots · IRIS - UNITN - prod
Checking a claim in this area?
We can run the same search on any method or claim. If nothing turns up, we will say so, and that proves nothing on its own.