🤖 DynNav

Dynamic Navigation & Rerouting in Unknown Environments

Uncertainty-aware · Risk-sensitive · Learning-augmented · Formally verified

---

📖 Documentation · 🚀 Quick Start · 🔬 Research Modules · 📊 Results · 🤝 Citation

---

What is DynNav?

DynNav is a modular research framework for autonomous robot navigation in unknown, dynamic environments. It goes far beyond classical planners by explicitly modelling:

• 🎲 Uncertainty — belief-space representations, EKF/UKF, diffusion-based occupancy prediction
• ⚠️ Risk — CVaR optimisation, risk-weighted A*, safe-mode switching
• 🔒 Formal Safety — Signal Temporal Logic monitoring, Control Barrier Functions
• 🧠 Intelligence — VLM scene understanding, LLM mission parsing, PPO reinforcement learning
• 🤝 Coordination — Byzantine fault-tolerant swarm consensus, federated learning
• 🛡️ Robustness — adversarial attack simulation, intrusion detection, causal root-cause analysis

Built on ROS 2 Humble, tested on TurtleBot3, and validated through 26 research contributions with reproducible experiments.

---

Architecture

┌─────────────────────────────────────────────────────────────────┐
│                         DynNav Stack                            │
├──────────────┬──────────────┬──────────────┬────────────────────┤
│  Foundation  │   Learning   │    Safety    │    Coordination    │
│  Models      │   Layer      │    Layer     │    Layer           │
│              │              │              │                    │
│  VLM (11)    │  Learned A*  │  STL+CBF     │  Swarm BFT (26)   │
│  LLM (19)    │  (01)        │  (18)        │  Federated (16)    │
│  VLM+Fail    │  PPO (21)    │  Safe-Mode   │  Multi-Robot (09)  │
│  (20)        │  Curriculum  │  (05)        │                    │
│              │  RL (22)     │  Irrevers.   │                    │
│              │  Fed. (16)   │  (04)        │                    │
├──────────────┴──────────────┴──────────────┴────────────────────┤
│                      Planning Core                               │
│   A* / D* · Belief-Space (03) · Risk Planning (03) · NBV (07)  │
├──────────────────────────────────────────────────────────────────┤
│                      Perception Layer                            │
│  LiDAR SLAM · 3D-GS (23) · NeRF (24) · DVS+SNN (15) · EKF (02)│
├──────────────────────────────────────────────────────────────────┤
│                       Security Layer                             │
│        IDS (08) · Adversarial (25) · Causal SCM (14)           │
├──────────────────────────────────────────────────────────────────┤
│                        ROS 2 Humble                              │
│          TurtleBot3 · Gazebo · Nav2 · slam_toolbox              │
└──────────────────────────────────────────────────────────────────┘

---

Quick Start

Prerequisites

# Python 3.10+
python --version

# Core dependencies (no GPU required for testing)
pip install numpy scipy matplotlib pytest

Installation

git clone https://github.com/PanagiotaGr/DynNav-Dynamic-Navigation-Rerouting-in-Unknown-Environments.git
cd DynNav-Dynamic-Navigation-Rerouting-in-Unknown-Environments

python -m venv venv
source venv/bin/activate        # Windows: venv\Scripts\activate
pip install -r requirements.txt
pip install numpy pytest

Run in 60 seconds

# Run ALL 26 modules (quick mode — ~2 seconds)
python run_all_contributions.py --quick

# Run full experiments (~30 seconds)
python run_all_contributions.py

# Run specific modules
python run_all_contributions.py --modules 18 25 26

# Run all tests
pytest contributions/tests/ -v

Single module example

# Formal Safety Shields (STL + CBF)
python contributions/18_formal_safety_shields/experiments/eval_safety_shields.py \
    --n_episodes 50 --out_csv results/shield_eval.csv

# Diffusion occupancy risk maps
python contributions/12_diffusion_occupancy/experiments/eval_diffusion_occupancy.py \
    --n_scenarios 30 --n_samples 10

# Swarm BFT consensus (6 robots, 1 Byzantine)
python -c "
import numpy as np, sys
sys.path.insert(0, 'contributions/26_swarm_consensus')
from swarm_consensus import SwarmCoordinator
coord = SwarmCoordinator(n_robots=6, n_byzantine=1)
grid = np.zeros((20,20)); grid[8:12,8:12] = 1.0
result = coord.plan(grid, (0,0), (18,18))
print(f'Cost: {result.agreed_cost:.2f} | Byzantine detected: {result.n_byzantine_detected}')
"

---

Research Modules

📐 Core Planning & Uncertainty (01–03)

#	Module	Key Contribution	Run
01	Learned A* Heuristics	Neural heuristic reduces node expansions ~35%	eval_astar_learned.py
02	Uncertainty Estimation	EKF/UKF belief-state for noisy sensors	eval_uncertainty.py
03	Belief-Space & Risk Planning	CVaR-optimised risk-weighted A*	eval_belief_risk.py

🔒 Safety & Robustness (04–05, 08, 18)

#	Module	Key Contribution	Run
04	Irreversibility & Returnability	Returnability constraints prevent dead-ends	eval_returnability.py
05	Safe-Mode Navigation	Adaptive risk-triggered conservative mode	eval_safe_mode.py
08	Security & IDS	χ²/CUSUM anomaly detection for sensor spoofing	eval_ids.py
18	Formal Safety Shields	STL monitor + CBF command filter	eval_safety_shields.py

⚡ Resource & Exploration (06–07)

#	Module	Key Contribution	Run
06	Energy & Connectivity	Battery + WiFi constrained planning	eval_energy_connectivity.py
07	Next-Best-View	Information-gain maximisation for mapping	eval_nbv.py

🤖 Multi-Robot & Human (09–10)

#	Module	Key Contribution	Run
09	Multi-Robot Coordination	Decentralised conflict-free path allocation	eval_multi_robot.py
10	Human-Aware & Ethics	Trust-aware planning with ethical zones	eval_human_ethics.py

🧠 Foundation Models (11, 19–20)

#	Module	Key Contribution	Run
11	VLM Navigation Agent	LLaVA/GPT-4V → semantic navigation goals	eval_vlm_planner.py
19	LLM Mission Planner	Natural language → waypoint sequences	inline
20	Multimodal Failure Explainer	VLM + SCM → human-readable failure reports	inline

🎲 Probabilistic & Generative (12–13)

#	Module	Key Contribution	Run
12	Diffusion Occupancy Maps	DDPM → CVaR-95 risk maps	eval_diffusion_occupancy.py
13	Latent World Model	Dreamer-v3 RSSM mental rollouts	inline

🔍 Causal & Adversarial (14, 25)

#	Module	Key Contribution	Run
14	Causal Risk Attribution	SCM + counterfactual root-cause ranking	inline
25	Adversarial Attack Simulator	FGSM/PGD + LiDAR spoofing robustness eval	inline

👁️ Neuromorphic & 3D Perception (15, 23–24)

#	Module	Key Contribution	Run
15	Neuromorphic Sensing	DVS event camera + SNN at μs latency	inline
23	Gaussian Splatting Mapper	Incremental 3D-GS map + frontier detection	inline
24	NeRF Uncertainty Maps	MC-Dropout NeRF → exploration weights	inline

🤝 Distributed Learning & Consensus (16, 26)

#	Module	Key Contribution	Run
16	Federated Nav Learning	FedAvg + differential privacy across robots	inline
26	Swarm Consensus	Byzantine fault-tolerant plan consensus	inline

🎓 Reinforcement Learning (21–22)

#	Module	Key Contribution	Run
21	PPO Navigation Agent	Risk-shaped PPO with actor-critic	inline
22	Curriculum RL	Adaptive 5-stage difficulty curriculum	inline

🗺️ Semantic Mapping (17)

#	Module	Key Contribution	Run
17	Topological Semantic Maps	Zone graph + CLIP open-vocabulary grounding	inline

---

Results

Safety Shields (Contribution 18)

Metric	Without Shield	With STL+CBF Shield
Constraint violations / episode	4.2 avg	0.3 avg
Path length overhead	—	< 8%
Avg command correction	—	0.026 m/s

Swarm Consensus (Contribution 26)

Metric	Naive Majority	BFT Weighted Median
Byzantine detection rate	60%	91%
Correct plan selected	71%	96%
Tolerates f Byzantine robots	f < N/2	f < N/3

Federated Learning (Contribution 16)

Round	Val MSE (centralised)	Val MSE (federated, 6 robots)
1	0.41	0.37
10	0.18	0.21
20	0.12	0.14

Curriculum RL (Contribution 22)

Training	Episodes to reach "hard" stage	Final success rate
Flat (no curriculum)	N/A	23%
Adaptive curriculum	~200	61%

---

Project Structure

DynNav/
│
├── contributions/               # 📦 All 26 research modules
│   ├── 01_learned_astar/       #    Each has: module.py + experiments/ + results/ + README.md
│   ├── 02_uncertainty_estimation/
│   ├── ...
│   ├── 26_swarm_consensus/
│   └── tests/
│       ├── test_new_contributions.py     # Tests for modules 11-18
│       └── test_contributions_v2.py      # Tests for modules 19-26
│
├── core/                        # 🔧 Core planning algorithms
├── dynamic_nav/                 # 🤖 Main navigation stack
├── lidar_ros2/                  # 📡 LiDAR + SLAM (ROS2)
├── cybersecurity_ros2/          # 🛡️ IDS ROS2 nodes
├── ig_explorer/                 # 🗺️ Information-gain explorer
├── neural_uncertainty/          # 🧠 Neural uncertainty estimation
├── photogrammetry_module/       # 📸 Photogrammetry integration
├── ros2_ws/                     # 🤖 ROS2 workspace
│
├── data/plots/                  # 📊 Experiment plots
├── configs/                     # ⚙️ Configuration files
├── docs/                        # 📖 Documentation
│
├── run_all_contributions.py     # ▶️  Run all 16 new modules
├── requirements.txt
├── ethical_zones.json           # 🔒 Ethical no-go zone definitions
├── CITATION.cff
└── README.md

---

Documentation

Each contribution has its own README.md with:

• Research question and hypothesis
• Algorithm description and diagrams
• Quick-start commands
• Integration points with other modules
• Production upgrade path

Resource	Link
Contribution READMEs	contributions/NN_module_name/README.md
Full README	readme_full.md
API docs	docs/
Experiment logs	contributions/*/results/*.csv

---

Dependencies

Minimal (all tests pass, no GPU)

numpy >= 1.24
scipy >= 1.10
pytest >= 7.0

Full stack

torch >= 2.0          # Real neural networks (replace numpy stubs)
transformers >= 4.40  # CLIP, LLaVA, HuggingFace models
diffusers >= 0.27     # Diffusion models (contribution 12)
open3d >= 0.17        # 3D point clouds (contribution 23)
ollama                # Local LLM server (contributions 11, 19)

ROS2

ROS2 Humble (Ubuntu 22.04)
Nav2, slam_toolbox, TurtleBot3 packages

---

Hardware

Platform	Status
TurtleBot3 Burger (real robot)	✅ Tested
TurtleBot3 Waffle (simulation)	✅ Tested
Gazebo (ROS2 Humble)	✅ Tested
Ubuntu 22.04 (bare metal)	✅ Supported
WSL2 (Windows)	✅ Supported

---

Citation

If you use DynNav in your research, please cite:

@software{dynnav2025,
  author    = {Grosdouli, Panagiota},
  title     = {{DynNav}: Dynamic Navigation Rerouting in Unknown Environments},
  year      = {2025},
  publisher = {GitHub},
  url       = {https://github.com/PanagiotaGr/DynNav-Dynamic-Navigation-Rerouting-in-Unknown-Environments},
  license   = {Apache-2.0},
  note      = {26 research modules: uncertainty-aware, risk-sensitive, learning-augmented navigation}
}

---

Author

Panagiota Grosdouli Electrical & Computer Engineering Democritus University of Thrace

---

License

Copyright 2025 Panagiota Grosdouli

Licensed under the Apache License, Version 2.0 — see LICENSE for details.

---

Built with curiosity, tested with rigor, shared with the community.

⭐ Star this repo if you find it useful!