Silent Sentry — Autonomous UGV Navigation in Contested Environments

---

Abstract

Silent Sentry presents a multi-layer autonomous navigation stack for Unmanned Ground Vehicles (UGVs) operating in unstructured, GPS-degraded desert corridors. The system integrates:

• EMCON Controller — Emission-Control-aware command arbitration that suppresses unnecessary RF/sensor emissions during adversarial exposure windows
• SBLP Planner — Scenario-Based Local Planner that switches trajectory primitives based on terrain class (sand dune, rock field, open corridor)
• VLM Costmap — Vision-Language Model costmap layer that annotates traversability using zero-shot scene understanding (no fine tuning required)

Experiments are conducted in a photorealistic Gazebo Harmonic simulation of the Thar Desert corridor (1735 × 4144 m heightmap, 187 vegetation models).

---

Architecture

/cmd_vel (Twist)
      │
      ▼
┌─────────────┐     terrain class     ┌──────────────┐
│ VLM Costmap │ ──────────────────▶  │ SBLP Planner │
│  (vlm_costmap)│                    │ (sblp_planner)│
└─────────────┘                      └──────┬───────┘
                                            │ waypoints
                                            ▼
                                   ┌─────────────────┐
                                   │ EMCON Controller │
                                   │ (emcon_controller)│
                                   └────────┬────────┘
                                            │ /forward_velocity_controller/commands
                                            ▼
                                        [autobot]

---

Repository Layout

silent_sentry_ros2/
├── README.md
├── .gitignore
├── docker/
│   ├── Dockerfile            # ROS 2 Jazzy + Gazebo Harmonic base image
│   └── docker-compose.yml    # Multi-agent network emulation (3 UGVs)
├── paper_data/
│   ├── plotting_scripts/     # matplotlib scripts to reproduce all paper figures
│   └── final_graphs/         # PNGs used in the IEEE submission
├── src/
│   ├── emcon_controller/     # EMCON-aware command arbitration node
│   ├── sblp_planner/         # Scenario-based local planner
│   └── vlm_costmap/          # VLM traversability costmap layer
└── world/
    └── thar_corridor.sdf     # Gazebo world (references local .tif heightmap)

---

Dependencies

Package	Version
ROS 2	Jazzy
Gazebo	Harmonic 8.10+
Python	3.12
OpenCV	4.x
Transformers (HuggingFace)	≥ 4.40
robot_localization	Jazzy

---

Build & Run

# 1. Clone
git clone https://github.com/YOUR_ORG/silent_sentry_ros2.git
cd silent_sentry_ros2

# 2. Build
colcon build --symlink-install
source install/setup.bash

# 3. Launch full stack
ros2 launch emcon_controller silent_sentry.launch.xml

Docker (multi-agent)

cd docker
docker-compose up --build

---

Reproducing Paper Figures

cd paper_data/plotting_scripts
python3 plot_traversability.py
python3 plot_emcon_timeline.py
python3 plot_sblp_scenario_switch.py

Output PNGs will appear in paper_data/final_graphs/.

---