Chapter Four · failure evidence
What Autonomous Systems Control got wrong, from 16 dissertations
The records describe technical failures and design rejections across diverse autonomous vehicle and robotic control systems. Key challenges include unmodeled dynamics, severe environmental disturbances, optimization infeasibility, and inappropriate architectural assumptions. These records come from PhD theses at 10 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.
Severe environmental disturbances exceed vehicle control and compensation authority
Autonomous underwater vehicles maneuvering in waves failed because environmental wave perturbations dominated vehicle dynamics and exceeded compensation envelope capabilities. Similarly, fixed-wing unmanned aerial vehicles lost flight stability in turbulent ship airwakes because crosswind gusts created aerodynamic side-forces exceeding lateral control authority and vehicle weight.
Tried and failed
multi-fidelity active learning for hydrodynamic compensation applied to autonomous underwater vehicle maneuvering in waves. Outcome: no signal. Reason: environmental wave-induced perturbations dominated the vehicle dynamics, exceeding compensation envelope capabilities
Tried and failed
fixed-wing UAV autonomous flight control applied to turbulent ship airwake environments. Outcome: unstable. Reason: aerodynamic side-force from crosswind gusts exceeded lateral control authority and vehicle weight
A Dynamic Model of Unpiloted Aerial Systems for Complex Ship Airwake Environments · Georgia Tech
Simplified dynamic formulations fail under aggressive high-speed maneuvering
Simultaneous planning and tracking nonlinear model predictive control became unstable during high-speed cornering due to decoupled lateral dynamics and slow yaw rate response. In autonomous racing, a kinematic model with Gaussian process corrections was outperformed by model predictive control utilizing full dynamic model knowledge prior to multi-lap updates.
Tried and failed
simultaneous planning and tracking nonlinear model predictive control applied to autonomous vehicle high-speed cornering. Outcome: unstable. Reason: decoupled lateral dynamics and slow yaw rate response during aggressive cornering
A non-linear optimal strategy for simultaneous planning and tracking of autonomous vehicles · Cranfield
Lost to a baseline
In autonomous racing, BestCase MPC with full dynamic model knowledge beat BayesRace GP-corrected kinematic model prior to multi-lap model updates
Learning-based and adaptive controllers exhibit sustained oscillations and limit cycles
Omitting hardware actuator rate limits during simulation-to-real reinforcement learning produced real-world heading oscillations and rotational overshoot in marine vessel control. Additionally, non-parametric reproducing kernel Hilbert space adaptive control exhibited steady-state limit cycle oscillations in pitch and depth during underwater vehicle tracking.
Tried and failed
sim-to-real reinforcement learning omitting actuator rate limits applied to autonomous marine vessel control. Outcome: unstable. Reason: omitting hardware rate limits in simulation caused real-world heading oscillations and rotational overshoot
Autonomous System for Identifying and Capturing Floating Waste · Georgia Tech
Tried and failed
reproducing kernel hilbert space adaptive control applied to autonomous underwater vehicle trajectory tracking. Outcome: unstable. Reason: baseline non-parametric adaptive control exhibited steady-state limit cycle oscillations in pitch and depth
Data-Driven, Non-Parametric Model Reference Adaptive Control Methods for Autonomous Underwater Vehicles · Virginia Tech
Left open by the authors
Problems the authors named and did not get to.
Left open
Implement and evaluate PPO, DDPG, and model predictive control algorithms within the PRAECEPTION autonomous driving perception framework using CARLA. Blocker: None
Assessing and enhancing system-level safety in learning-enabled cyber-physical systems · Iowa State
Left open
Develop pattern recognition algorithms to dynamically switch autonomous vehicle longitudinal MPC to strict string stability upon detecting abnormal leader behaviors. Blocker: None
Longitudinal Control for Self-driving Cars with Traffic Flow Considerations: Theory, Design, and Experiments · Georgia Tech
Left open
Implement a real-time Model Predictive Control (MPC) feedback controller for autonomous vehicle trajectory tracking minimizing motion sickness dose values. Blocker: None
Control for motion sickness minimisation in autonomous vehicles. · Cranfield
Left open
Quantify causal relationships and magnitudes of personalization embedding shifts along the aggressive gradient to optimize autonomous vehicle driving style. Blocker: None
Data-driven personalization techniques to account for heterogeneity in human-machine interaction · Georgia Tech
Left open
Implement multi-step planning and longer prediction horizons by combining Model Predictive Control with Control Barrier Functions for cooperative autonomous vehicles. Blocker: None
Left open
Incorporate driverless program data and operational covariates into autonomous vehicle recurrent event reliability models. Blocker: Access to proprietary driverless vehicle fleet program operational and telemetry data
Time-to-Event Modeling with Bayesian Perspectives and Applications in Reliability of Artificial Intelligence Systems · Virginia Tech
Left open
Develop a model-free reinforcement learning controller for connected and autonomous vehicles to determine optimal speed reductions as mobile traffic controllers. Blocker: None
Modelling mixed traffic flow of autonomous vehicles and human-driven vehicles · Imperial
Left open
Develop an autonomous robot self-exploration routine to estimate task impedance and detect resistance/condition changes during articulated manipulation. Blocker: Requires a physical robotic manipulation setup with force/torque sensing to interact with articulated physical objects
Remote robotic manipulation task execution using affordance primitives · UT Austin
Left open
Deploy VoI-guided AUV planning algorithms under acoustic communication and bandwidth constraints during real oceanographic deployments. Blocker: Requires physical Autonomous Underwater Vehicles (AUVs), marine acoustic modems, and access to oceanographic research cruises
Balancing exploration and exploitation: task-targeted exploration for scientific decision-making · Woods Hole
Left open
Adapt VoI and open-loop macro-action execution for bandwidth-limited acoustic modems and deploy on oceanographic AUVs like SENTRY. Blocker: Requires access to physical oceanographic autonomous underwater vehicles (AUVs) and acoustic modem hardware for real-world deployment
Balancing Exploration and Exploitation: Task-Targeted Exploration for Scientific Decision-Making · MIT
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.