Electrical and Computer Engineering · Reliable Autonomous Systems · Uncertainty-Aware AI
I study how autonomous systems can remain reliable when perception is noisy, localization degrades, environments change, and predictive models become uncertain.
My long-term objective is to develop robotic systems that can:
- represent and propagate uncertainty across the autonomy stack;
- evaluate the health of their own perception and state estimates;
- anticipate failure before it becomes safety critical;
- adapt planning and control according to risk and recoverability;
- explain why a protective or recovery action was selected;
- produce reproducible evidence rather than unsupported performance claims.
Central research question
How can an autonomous system reason about what it does not know and convert that reasoning into safer real-time behaviour?
|
Failure-aware visual–inertial autonomy with real feature tracking, IMU preintegration, an error-state EKF, calibrated risk estimation, domain-shift monitoring, navigation shielding, and recovery actions. Focus: VIO · estimator introspection · failure prediction · runtime protection |
Adaptive fusion of heterogeneous sensing modalities using reliability estimation, covariance adaptation, innovation consistency, and failure-aware weighting. Focus: SLAM · sensor fusion · reliability · uncertainty propagation |
|
An architecture for transforming perceptual, state, environmental, and predictive uncertainty into risk-aware action selection and interpretable safety decisions. Focus: uncertainty reasoning · risk-sensitive autonomy · explainability |
Dynamic navigation and rerouting in partially observed environments with uncertainty-aware planning, recoverability analysis, and runtime replanning. Focus: planning · navigation · replanning · recoverability |
|
Interpretable trajectory prediction and interaction-risk reasoning for pedestrians, cyclists, vehicles, and intelligent intersections. Focus: VRU safety · trajectory prediction · TTC/CPA · interaction graphs |
AI-assisted research for underwater wireless optical communication, digital twins, adaptive link optimisation, and uncertainty-aware system analysis. Focus: underwater communication · digital twins · adaptive optimisation |
|
Research components for navigation decisions that account for uncertain maps, state estimates, future motion, and downstream safety risk. Focus: uncertainty-aware planning · safety supervision |
A simulation-oriented project for modelling race dynamics, strategy, uncertainty, and performance evolution. Focus: simulation · modelling · strategy analysis |
| Area | Methods and tools |
|---|---|
| Robot perception | OpenCV, visual tracking, semantic segmentation, trajectory prediction |
| State estimation | Kalman filtering, error-state EKF, IMU preintegration, VIO, SLAM |
| Uncertainty | covariance analysis, NIS/NEES, calibration, conformal prediction, domain shift |
| Planning and safety | risk-sensitive planning, dynamic replanning, runtime shielding, recovery policies |
| Machine learning | PyTorch, scikit-learn, interpretable baselines, uncertainty-aware models |
| Simulation | ROS 2, Gazebo, CARLA, SUMO, deterministic synthetic experiments |
| Engineering | Python, C/C++, Java, TypeScript, testing, CI, reproducible artifacts |
- Evidence before claims. Synthetic, dataset, simulator, and hardware evidence must remain clearly separated.
- Uncertainty must be evaluated. Confidence visualisation alone is not sufficient; calibration and estimator consistency matter.
- Failure is part of the system model. Reliable autonomy requires detection, protection, and recovery—not only nominal accuracy.
- Reproducibility is a research result. Seeds, configurations, commands, versions, and output artifacts should be recorded.
- Safety requires layered validation. Research prototypes are not deployment-ready safety systems.
Perception reliability
↓
Uncertainty-aware state estimation
↓
Failure and domain-shift prediction
↓
Risk-sensitive planning
↓
Runtime shielding and recovery
↓
Simulation, public datasets, and hardware validation
I aim to pursue doctoral research in robotics and autonomous systems, focusing on reliable perception, uncertainty-aware localization, risk-sensitive planning, and runtime safety for intelligent robotic systems.
- Email: panagros1@ee.duth.gr
- LinkedIn: Panagiota Grosdouli
- Research portfolio: panagiota-research-portfolio.vercel.app
- ResearchGate: Panagiota Grosdouli
Repositories contain projects at different maturity levels. Each project README should be treated as the source of truth for its implementation status and validation boundary.