Manual monitoring in hazardous environments is unsafe, time-consuming, and inefficient
Existing systems are expensive, fixed, and require continuous human control
There is a need for a mobile automated system for remote surveillance and data collection
The project proposes a low-cost UGV with wireless control, sensor-based navigation, and real-time monitoring
Current state-of-the-art achieves sub-10cm localization error in GPS-denied scenarios and ~97% terrain classification accuracy in complex environments like snow-covered or wooded terrains.
Critical bottlenecks remain in handling ultra-dynamic obstacles with low latency, distinguishing deformable vegetation from solid hazards, and maintaining communication mesh-stability in remote field deployments.
Our project addresses the identified gaps by integrating adaptive terrain-aware path planning with a decentralized sensor fusion framework, aiming to improve UGV survival rates in unstructured off-road environments.
"The convergence of Deep RL and Heterogeneous Multi-Robot Collaboration represents the next frontier in achieving true Level 5 autonomy for tactical and agricultural UGVs in the world's most challenging outdoor environments."
Develop cost-effective full-stack autonomous vehicle using limited sensor suite for unconstrained environments.
Four-layer architecture: Sensors → Perception → Planning → Control; Prior map-based localization + sensor fusion + rule/state-machine planning + PID/MPC control.
32-channel Velodyne LiDAR, 2 FLIR cameras, GNSS+IMU (Novatel), CAN bus, Raspberry Pi, server/laptop.
KITTI dataset (training), Real-world campus/unconstrained outdoor environment.
3D NDT localization, YOLOv3 (camera detection), PointPillars (LiDAR detection), IMM-UK-JPDAF tracking, A* path planning, State machine, PID & MPC control, Pure Pursuit.
Accurate localization (low RMSE vs GNSS), reliable object detection, successful obstacle avoidance/stopping, validated autonomous taxi service.
Limited sensor diversity, dependence on prior maps, proprietary vehicle control, constrained speed/testing scope.
Layered architecture design, NDT localization concept, LiDAR-camera fusion strategy, state-machine planning logic, PID/MPC control framework.
Enhance robustness via multi-sensor fusion (GPS+LiDAR+Visual SLAM), dynamic mapping, improved obstacle prediction, lightweight real-time models.
Reliable UAV navigation & localization in GPS-denied, unknown environments.
Air–ground collaborative perception; BEV image stitching + object detection + mobile UWB anchor trilateration + improved D* Lite planning.
DJI RoboMaster TT UAVs, downward cameras, UWB modules (Nooploop LinkTrack), laser height sensor, Wi-Fi network.
Custom BEV image dataset; real-world outdoor experiments; GPS-denied scenario.
SuperPoint + SuperGlue (stitching), YOLOv5 (object detection), UWB trilateration, Improved D* Lite path planning.
Real-time global 3D mapping; <10 cm localization error; fast path planning (milliseconds); successful collision-free navigation.
BEV coverage limitations; performance degradation under severe occlusion; global planner latency for highly dynamic obstacles.
Air–ground system design, BEV mapping concept, UWB localization strategy, D* Lite planning framework.
Integrate Visual SLAM + LiDAR fusion; DRL-based planner; dynamic obstacle prediction.
Reliable obstacle detection and object recognition for UGV navigation using low-cost sensors.
Neural-network-based object recognition using ultrasonic sensor data and MATLAB training.
5× SRF05 Ultrasonic Sensors, Arduino-Galileo controller, IMU sensors, DC gear motors.
Experimental test-rig + autonomous rover tests; ~3300 sensor samples collected.
Feed-forward Artificial Neural Network (MATLAB Neural Network Toolbox).
Object recognition accuracy ≈77% in preliminary tests; ~99% accuracy in real rover experiments.
Sensor interference, limited object types, small experimental environment.
Ultrasonic-based obstacle detection, neural network object classification approach.
Add LiDAR + camera-based perception with SLAM and advanced path planning.
Improving maneuverability, stability, obstacle negotiation, and autonomy of a multifunctional autonomous UGV.
Mathematical modeling + MATLAB & RecurDyn simulations + prototype testing + modular design integration.
GNSS, IMU, LiDAR, Ultrasonic sensors, Infrared sensors, Cameras, NVIDIA Jetson Xavier.
Simulation environments + Controlled real-world prototype experiments (slopes, steps, ditches).
Kinematic & dynamic modeling, Traction force modeling, PID control, Bug 2 obstacle avoidance.
Max climb angle ≈ 30°, obstacle height ≈ 200 mm, ditch crossing ≈ 400 mm; simulation & prototype alignment.
Idealized assumptions (constant friction), limited test scenarios, model–prototype discrepancies.
Traction modeling approach, obstacle analysis methods, PID stabilization strategy, modular platform concept.
Dynamic terrain modeling + adaptive control + sensor fusion (LiDAR + Visual SLAM + GPS).
Reliable navigation & localization in snow-covered, low-visibility environments.
UAV-guided UGV navigation using road segmentation + sensor fusion + hierarchical planning + message-queue communication.
UAV: RGB-D Camera, GPS, IMU | UGV: IMU (minimal sensors)
Custom UAV aerial dataset + Snow-covered simulated environment (Gazebo/ROS)
U-Net (road segmentation), Kalman Filter (fusion), RTAB-Map (SLAM), Frontier Exploration, YOLOv5 (UGV tracking), Pure Pursuit Controller, TEB + OMPL planners.
95.9% segmentation accuracy, 94% route success, <5% localization error, reduced path deviation (~12%), improved navigation accuracy.
Simulation-based validation, RGB-D dependency, limited real-world testing.
UAV-UGV collaboration strategy, segmentation-assisted planning, hierarchical navigation, tracking + Pure Pursuit control.
LiDAR + Visual SLAM fusion, dynamic terrain adaptation, RTK-GPS / GPS-denied robustness.
Safe & efficient UGV navigation in complex off-road environments considering diverse risk factors.
Global–local planning framework integrating APF-based risk assessment + Coarse2fine A* + deterministic sampling local planning.
LiDAR, Cameras, Perception & Mapping modules (real-world UGV deployment).
Simulated uncertainty maps + Real-world off-road experiments.
Artificial Potential Field (APF), Coarse2fine A*, Deterministic Sampling, Polynomial Trajectories, QP Smoothing.
~30% safer paths vs baseline; real-time performance; successful obstacle avoidance.
Model assumptions, limited terrain variability, computational dependency.
Risk-aware planning strategy, Coarse2fine A*, APF safety modeling.
Adaptive terrain learning + sensor fusion (LiDAR + Visual SLAM + GPS).
Design & development of manual/autonomous UAGV for safe navigation & military applications.
Integration of sensors + GPS navigation + Image processing + Computer vision + LabVIEW-based control.
Camera, Ultrasonic sensors, GPS, Digital compass, IR sensors, Zigbee, DC gear motors, Arduino, LiPo battery.
Image processing, Obstacle detection, Shortest path (Dijkstra), GPS-based navigation.
Successful manual & autonomous navigation; effective obstacle detection & remote control.
Sensor integration strategy, GPS navigation concept, obstacle detection framework.
Improve UGV navigation accuracy, responsiveness, and latency using IoT & sensor fusion.
IoT-based modular control system with edge computing, real-time sensor fusion, adaptive decision-making.
Sensor fusion, path planning, obstacle avoidance, machine learning models, edge processing.
GPS (RTK), IMU (9-axis), 360° LiDAR, ARM Cortex-M7 MCU, wireless/CAN communication.
Improved navigation accuracy, reduced latency, enhanced responsiveness.
IoT modular architecture, edge computing strategy, GPS-IMU-LiDAR fusion concept.
Comprehensive review of modular UGV advancements, design challenges, and application trends.
Literature survey (2016–2021); analysis of design parameters, control architectures, path planning, sensors, power systems.
Review Paper (LiDAR, Cameras, Ultrasonic, GPS, IMU discussed).
Global vs Local planning, Fuzzy Logic, ANN, MPC, Genetic Algorithms, SLAM.
Identifies trends: modularity benefits, sensor-driven autonomy, AI-based navigation improvements.
Modular design insights, sensor selection strategy, navigation & control taxonomy.
Short-range limitations of Bluetooth-based surveillance robots & need for safer long-range control.
Wi-Fi controlled UGV using NodeMCU (ESP8266), Android app interface, real-time video streaming, gas sensing.
NodeMCU (ESP8266), Wireless Camera, MQ-135 Gas Sensor, Motor Driver (L298N), DC Motors.
Wireless control logic, sensor monitoring (no complex AI).
Reliable Wi-Fi control, continuous video transmission, real-time air quality monitoring.
Wi-Fi control architecture, NodeMCU integration, sensor interfacing strategy.
Autonomous charging of UGV without human intervention and without fiducial markers.
YOLO-World, A*, DWA, DBSCAN, WPCA, Inverse Kinematics.
6-DoF manipulator, TurtleBot UGV, Intel RealSense D435i, LDS-02 LiDAR, NVIDIA Jetson Xavier NX.
90% docking success rate; average total time = 34.4 s.
Zero-shot detection for unseen objects; Marker-free docking; 6-DoF pose estimation pipeline.
Extend to dynamic targets; multi-robot coordination; improve alignment precision.
Fixed environmental sensors cannot capture spatial variations in indoor environmental quality (IEQ). Static sensors create undetected zones and spatial inefficiency.
Systematic literature review of 30 selected studies on UGV-based mobile sensing. Studies categorized into UGV structure, monitoring & data capture, data analysis, and future directions.
Mobile robots (mostly TurtleBot), sensors for temperature, humidity, CO₂, VOCs, PM2.5, illuminance, sound, localization sensors (LiDAR, cameras, encoders).
Experiments mostly in static indoor environments such as universities, labs, factories, and schools.
SLAM, Kalman Filter, interpolation (Kriging, cubic, nearest neighbor), machine learning models such as SVR, KNN, Linear Regression, Random Forest, Decision Trees.
UGV sensing provides high-resolution spatial-temporal environmental data, detects pollutant hotspots, improves IEQ mapping, and can reduce building energy consumption by ~3%.
Sensor response time (SRT) errors, localization challenges, limited real-world testing, limited validation methods, difficulty handling dynamic human occupancy.
Use mobile robots with environmental sensors to collect spatial environmental data; combine with machine learning models to predict indoor environmental conditions.
Integrate 3D indoor mapping, improved localization, optimized robot velocity for sensing, and integration with Building Automation Systems (BAS).
Robots operating in underground mines or disaster environments often lose communication due to lack of network infrastructure. This prevents real-time data transfer during search and rescue missions.
Developed a UGV-mounted system that deploys wireless communication nodes (breadcrumbs) to create a temporary mesh network while the robot moves through the environment.
UGV platform (Clearpath Husky A200), ESP32 microcontrollers, servo motors, accelerometer sensors, N20 motor with lead screw, Wi-Fi antennas.
Tested in uneven terrain environments simulating subterranean mining or SAR scenarios.
ESP-NOW communication topology, accelerometer-based orientation control, servo control loop.
System successfully deployed nodes and created communication extension capability; servo-actuated node self-righting time ≈1.9 s and gravity-gimbal node ≈2.86 s.
Prototype-level system, limited environmental testing, mesh network performance not fully evaluated, communication modules may need improvement for harsh environments.
Deployable mesh communication nodes for UGV exploration; modular node deployment architecture; self-righting communication node concept.
Add automatic node placement optimization using signal strength or AI-based positioning, integrate long-range communication (LoRa / UWB).
Precise communication for autonomous systems in agriculture often relies on 4G/5G, which is limited in vast/inaccessible terrains, hindering data relay to servers.
Development of a P2P ad hoc network using UAVs as mobile access points. Integrates Smell Agent Optimization (SAO-P2PS) for dynamic peer selection under fluctuating link conditions.
UGV (environmental sensors for soil/humidity), UAV (mobile AP), Wi-Fi/Bluetooth communication modules.
Vast agricultural terrains with limited or zero 4G/5G coverage.
Smell Agent Optimization-based Peer-to-Peer Selection (SAO-P2PS) algorithm.
SAO-P2PS achieves 25-35% higher throughput and 40-60% better energy efficiency compared to Random or Nearest-Neighbor selection.
Communicational network instability, computational constraints, and environmental adaptability in extreme weather.
Cost-efficient ad hoc communication deployment; minimized reliance on infrastructure; real-time data collection framework.
Circumvents the need for traditional 4G/5G via adaptive UAV relay; optimized P2P selection for energy/throughput.
Traditional navigation (GNSS/Vision) fails in dense row crops due to signal attenuation, multi-path errors, and leaf-induced visual clutter.
Fusion of tactile mechanical probes with LiDAR. Filters rigid obstacles (stalks) from flexible features (leaves) to navigate without visual input.
Mechanical probe with rotary encoder, 2D/3D LiDAR (Hokuyo UTM-30LX/Velodyne VLP-16), IMU, Wheel encoders.
Densely planted sorghum and corn fields with narrow row spacing and cluttered canopies.
Obstacle mapping, heuristic row-line estimation, linear least squares for within-row distance, Model Predictive Control (MPC).
97% obstacle detection accuracy; 4cm global positioning accuracy; successfully navigated >100m in simulation and 30m in real fields.
High computational complexity of MPC solvers; reliance on rigid vs flexible feature differentiation; sensitivity to extreme row curvature.
Vision-independent navigation framework; tactile-LiDAR fusion strategy; robust under-canopy operation.
Explicitly differentiates rigid (stalks) vs flexible (leaves) obstacles; vision-independent stability in dense forest/crop environments.
Autonomous microgrids require physical cable connections between distributed energy nodes, but manual cable deployment is inefficient and dangerous in complex environments.
Designed a custom Adjustable Cable Management Mechanism (ACMM) that automatically deploys and retracts cable while the UGV moves, preventing slack or entanglement.
UGV platform, NEMA 24 stepper motor, Arduino Uno, rotary encoder, motor driver (DMT542T), cable reel system.
Outdoor terrain demonstration with a UGV connecting power source and load nodes.
Feedforward + PID feedback control based on UGV velocity and cable angle estimation.
Precise cable control ±1 inch over 70 ft cable, adjustable speed up to 3.3 ft/s, lightweight (~5 lbs system).
3D printed components may lack durability for long-term use; limited real-world testing; requires proper calibration.
Cable deployment mechanism, velocity-based cable control, Arduino + stepper motor implementation.
Use AI-based path prediction to optimize cable route, integrate autonomous cable obstacle detection.
UGVs face traction loss and irregular terrain; need robust trajectory tracking without complex derivatives.
Simplified 2nd-order SMC; Cascaded architecture (trajectory gen, pose estimation, LiDAR evasion, speed regulation).
RPLiDAR A2, GNSS, redundant IMUs, Magnetometer; 4WD differential-drive UGV.
Second-order Sliding Mode Control (SMC), GNSS-IMU pose fusion, LiDAR point cloud processing.
Tracking error ~0.08m; std. dev. of position/velocity low; robust to disturbances.
SMC for robust outdoor control; LiDAR-GNSS pose fusion; real-time evasion logic.
UGVs struggle with unstructured outdoor navigation due to unpredictable obstacles and terrain variations.
Hierarchical deep RL: Semantic perception + High-level RL policy + Low-level geometric controller.
RGB-D cameras, LiDAR; wheeled/tracked UGV; NVIDIA GPU.
HRL (PPO/DDPG), Semantic segmentation CNN (U-Net/DeepLab), Reward shaping.
86.7% trail-following success; low collision frequency; precise path tracking; superior vs. baselines.
Hierarchical RL architecture; semantic perception pipeline; reward shaping strategies.
Off-road navigation requires handling unstructured terrain traversability without extensive generic training data.
Deep RL with Actor-Critic; Curriculum Learning across 3 Gazebo scenarios; 3D LiDAR → virtual 2D traversability scanner.
3D LiDAR, Andabata UGV platform, onboard computing.
Actor-Critic RL, Random Forest traversability classifier.
Successful sim-to-real transfer; collision avoidance and goal reaching in natural terrain; accelerated convergence.
Traversability mapping from 3D LiDAR; curriculum learning; Actor-Critic RL.
Seamless integration of separate modules (perception, planning, control) for real-time outdoor navigation.
Three-module architecture: Line/obstacle detection, Path planning/tracking, Real-time motor control.
LiDAR scanner, monocular camera, motor control system hardware.
LiDAR-camera fusion, path planning within boundaries, motor control.
Successful real-time navigation within boundaries; demonstrated system integration performance.
Modular three-layer architecture; LiDAR-camera fusion pipeline; integration framework.
UGVs in GNSS-denied terrains need terrain-aware stealth navigation stack without GPS.
Hierarchical stack: vision-inertial mapping (VIO) → stealth planning → trajectory optimization → control.
Vision sensors, IMUs, Intel RealSense, Clearpath Husky A200.
Vision-Inertial Odometry (VIO), terrain segmentation, trajectory optimization, robust control.
Successful GNSS-denied navigation; 28Hz terrain segmentation; validated indoor/outdoor tracking.
GNSS-denied VIO pipeline; hierarchical autonomy stack; terrain-aware planning.
Precise real-time maneuvering across challenging outdoor terrains with dynamic obstacles and uncertainty.
Multi-modal sensor fusion + enhanced perception + memory-guided decision-making strategies.
LiDAR, Cameras, IMU, GNSS; high-performance onboard computing.
Deep learning fusion, memory-augmented navigation policies, adaptive trajectory optimization.
Superior real-time performance; robust decision-making; validated field performance.
Multi-modal sensor fusion pipeline; memory-guided navigation strategies; terrain classification.
Reliable water hazard detection for cross-country navigation; lack of standardized metrics.
Four image-space water detection algorithms fused into terrain classification; stereo ranging in 3D maps.
Stereo camera pairs (3 baselines), GDRS ladar, IMU, GPS; experimental XUVs.
Water detection fusion, stereo range fusion, temporal filtering in terrain maps, Kalman-filtered GPS/IMU.
Established detection range/accuracy metrics; ground truth within 0.5% positioning error.
Water detection fusion methodology; image vs. map space metrics; ground-truth validation protocol.
Collision-free global path planning in unstructured environments using satellite imagery without ground sensors.
Satellite image processing for path extraction + heuristic cost map generation + optimal planning algorithm.
Satellite imagery (offline/remote source).
Satellite image segmentation, heuristic cost functions, A*-variant optimal path search.
Generates paths 3-5x faster than traditional A*; 95%+ path optimality on unstructured terrains.
Satellite preprocessing as global planner; heuristic cost map generation module.
Manual forest nursery inventory and navigation is labor-intensive and inefficient. Existing navigation methods (GNSS, LiDAR) are expensive or unreliable in agricultural environments.
Developed an automatic guidance system for a UGV using stereo vision and depth image processing to detect nursery bed edges and guide the vehicle.
Skid-steering UGV, ZED 2i stereo camera, RTK-GNSS receiver, Jetson AGX Orin edge computer, LED lighting, OAK-D stereo cameras.
Tested in Webots simulation and real-world forest nursery fields with curved seedling beds.
Depth image processing, Canny edge detection, Hough line transform, proportional control steering algorithm.
Successfully completed 1422 m navigation without collisions, average RMS deviation ≈ 0.05 m from bed centerline.
Performance affected by lighting conditions, bed geometry, and soil conditions; stereo camera depth noise and detection failures at higher speed.
Stereo vision navigation, ROS-based robotic control architecture, depth-image processing for path detection.
Improve robustness using deep learning segmentation, adaptive controllers, and multi-sensor fusion (LiDAR + vision).
UGVs struggle with perception and traversability in unstructured environments like forests and mountains.
Presents TAERO platform; reviews perception challenges and AI solutions for unstructured operations.
LiDAR, cameras, radar, UWB radar for vegetation penetration; TAERO UGV platform.
Terrain classification (SPOC CNN), deep learning sensor interpretation, real-time path planning.
TAERO demonstrated for reconnaissance; UWB radar distinguishes grass/bushes from solid obstacles.
TAERO platform requirements as benchmark; UWB radar vegetation penetration concept.
Existing UAV datasets lack sensor synchronization, terrain awareness, and realistic environmental conditions, making localization research difficult.
Proposed OS-RFODG, a ROS2-based framework that integrates QGIS, Blender, Gazebo, PX4, and QGroundControl to generate synchronized multi-sensor UAV datasets with realistic terrain models.
LiDAR, GPS, IMU, RGB Camera, Barometer.
Simulation environment based on Gazebo + PX4 SITL, terrain from GeoTIFF elevation maps (Makkah region, Saudi Arabia).
Sensor fusion, Visual localization (SIFT feature matching), trajectory analysis, dataset benchmarking.
RMSE for trajectory elevation: 13.59–19.5 m; image localization precision up to 0.8964; visual localization error < 1.34 m.
Simulation-based dataset; elevation errors increase in complex terrain; environmental disturbances like wind/rain not fully modeled.
Framework pipeline for multi-sensor dataset generation, ROS2 synchronization techniques, terrain modeling pipeline using DEM/DSM + Blender.
Add real UAV flights + real sensor noise models, integrate AI-based localization, include dynamic environmental disturbances.
Autonomous navigation in rough forest terrains is difficult because robots must avoid obstacles, steep slopes, and unsafe paths while minimizing energy consumption.
Proposed Mechanical Effort Based Traversability (MEBT) framework that analyzes 3D point cloud terrain maps, computes terrain gradient and mechanical effort, and generates an optimal path on a traversability costmap.
UGV robot platform (Clearpath Husky), Velodyne LiDAR, IMU, wheel encoders.
Gazebo simulation scenarios and real forest dataset (Montmorency Forest, Canada).
Gradient computation, traversability analysis, mechanical effort estimation, A* global planning with ROS move_base.
Generated safer paths with lower mechanical effort and lower pitch risk; significantly improved map processing speed compared to previous mechanical-effort approaches.
System relies on pre-mapped point cloud environment; limited handling of dynamic obstacles; planning mainly focuses on global planning rather than full autonomy.
Traversability map generation, gradient-based terrain evaluation, mechanical-effort path planning concepts.
Add real-time perception + deep learning terrain classification + adaptive path planning using reinforcement learning or MPC.
| Product | Company | Application | Navigation / Sensing | Payload / Mobility | Approx. Price | Remarks |
|---|---|---|---|---|---|---|
| Husky A200 UGV | Clearpath Robotics | Outdoor R&D, inspection | ROS-ready, LiDAR/GPS/IMU | 75 kg, all-terrain | ₹16.8 lakh | Base price listed |
| Husky A300 UGV | Clearpath Robotics | Autonomous robotics | ROS 2 compatible | 100 kg, IP54 | Not disclosed | Commercial quote |
| RB-VOGUI | Robotnik | Inspection, logistics | Autonomous, sensor integration | Industrial base | Not disclosed | Vendor pricing |
| RB-FIQUS | Robotnik | Heavy industrial use | Modular autonomous system | Heavy payload | Not disclosed | Enterprise model |
| HK1000-V2 | SuperDroid Robots | Autonomous robot dev | GPS + LiDAR support | Heavy-duty terrain | ₹8–15+ lakh | Config-based |
| Proposed UGV | Academic Prototype | Navigation & surveillance | GPS L76x + IMU + LiDAR + ROS | Compact | ₹85k – ₹1.15 lakh | Student model |
| Feature | Existing UGVs | Proposed UGV |
|---|---|---|
| Cost | High | Low |
| Platform | Industrial | Academic |
| Outdoor Capability | Yes | Yes |
| Modularity | Medium–High | High |
| Replication | Difficult | Easy |
| Student Use | Limited | Strong |
| Sensor Fusion | Yes | Yes |
| Patent No. | Title | Year | Core Idea | Technologies Used |
|---|---|---|---|---|
| US12179737B2 | Unmanned Ground Vehicle and Method for Operating UGV | 2024 | Waypoint-based navigation with obstacle replanning | Waypoint Navigation, Obstacle Detection |
| US12269720B2 | Ground-Based Transport Vehicle | 2025 | Autonomous transport with minimal human input | Navigation Logic, Collision Avoidance |
| US10240930B2 | Sensor Fusion | 2019 | Combines multiple sensors for accurate localization | Sensor Fusion, Data Integration |
| US12019173B2 | Lane-Level Navigation using LiDAR & Signals | 2024 | High-precision navigation using LiDAR + signals | LiDAR Navigation, Localization |
| US11573090B2 | LiDAR and Remote Localization | 2023 | LiDAR-based mapping and navigation | LiDAR Mapping, Obstacle Detection |