projects
Research tracks, field prototypes, and robotics platforms from AVIS Lab.
From publications to working prototypes
AVIS Lab builds autonomous robots, intelligent sensing, and AI for field missions
The lab news and publication stream point to several concrete project tracks: UAVs for rescue and surveillance, UUVs and underwater multi-robot systems, robot coordination, SLAM localization, 3D sensing, robot vision, and lightweight AI models for recognition and sensing data. This page consolidates those tracks into the project areas AVIS Lab is actively developing.
- 6
- project tracks from news and papers
- UAV + UUV
- autonomous robot platforms
- AI + Sensors
- perception, localization, control
Flagship project
UAV and AI for firefighting and rescue support
This ministry-level project led by Dr. Pham Hoang Anh brings together the lab's core capabilities: UAV platforms, optical and thermal cameras, a ground station, digital maps, autonomous route planning, and AI-based scene understanding.
UAV, digital-map mission control, and AI scene perception
The system targets mission flight, camera-data collection, real-time visualization, fire-scene analysis, and human detection in observed areas. It connects flight-platform engineering, mission-control software, field datasets, and AI models for emergency response scenarios.
Read project newsActive project tracks
Consolidated from lab news, publications, and prototypes
The blocks below are not isolated titles. They are longer-running engineering and research tracks, each connecting hardware, software, simulation, datasets, and published research.
Flight platforms for surveillance, rescue, and event localization
The UAV track runs from quadrotor control and trajectory tracking to deep-learning obstacle avoidance and cooperative UAV swarms. The news stream connects this work to high-rise victim search, emergency response, acoustic explosion localization with ROS2/PX4, and digital-map mission control.
- Airframe integration, camera payloads, telemetry, ground station workflows, and flight logging.
- Waypoint missions, task monitoring, optical and thermal camera data collection.
- AI for obstacle avoidance, human detection, fire-scene analysis, and multi-UAV cooperation.
Coordinated underwater drones for hazardous environments
Publications from 2018 to 2023 show a sustained multi-UUV research line: coordination architecture, embedded control, formation tracking, collision avoidance, relative-localization estimation, and adaptive neural-network control for groups of low-cost underwater drones.
- Mission modeling, simulation, and multi-agent robot integration in open-source environments.
- Depth and heading control, formation tracking, collision avoidance, and underwater data collection.
- Coordination architectures spanning embedded systems, relative localization, and experimental validation.
Robot soccer, swarm perception, and cooperative decision-making
The RoboCup and multi-agent robot swarm publications show that AVIS Lab works beyond single-robot autonomy. This track covers perception, coordination, and decision-making for teams of robots, including hierarchical QMIX for soccer offense, the RoboCup Vision dataset, Dec-POMDP behavior design, and robot-human swarm perception.
- Perception pipelines for robot soccer, object detection, and shared visual datasets.
- Cooperative decision-making with probabilistic strategies, Dec-POMDP models, and reinforcement learning.
- Sensing and perceptual strategies for robot swarms and mixed robot-human collaboration.
Indoor localization, mapping, and tracking for mobile robots
Alongside UAV and UUV platforms, the lab maintains a localization and SLAM stack so autonomous systems can understand where they are. Prototypes use ORB-SLAM3, stereo cameras, LiDAR, and low-cost indoor localization methods for swarm robotics.
- Visual SLAM, visual-inertial SLAM, map management, loop closing, and trajectory tracking.
- Comparative analysis of low-cost indoor localization methods for robot swarms.
- Camera, IMU, and 3D sensor fusion to improve stability in low-texture or difficult lighting conditions.
LiDAR, stereo vision, and sensor fusion for real environments
The LiDAR and Stereo Vision projects form the sensing layer for autonomous robots. This track supports obstacle avoidance, 3D reconstruction, depth estimation, object detection, and downstream data for SLAM, navigation, and AI perception models.
- Laser ranging, point clouds, obstacle detection, and localization support when camera-only sensing is fragile.
- Stereo depth estimation, object detection, and geometric perception for mobile robots.
- Sensor-data preparation for simulation, algorithm evaluation, and multi-robot integration.
Image recognition, data enrichment, and sensing reconstruction
Another group of news items centers on lightweight AI models, vision backbones, and sensing-data reconstruction. This supports robot perception through image recognition, object detection, simulation-driven data enrichment, models for resource-limited devices, and compressed sensing for foot-pressure reconstruction.
- Lightweight backbones for image recognition, grouped dilation, and tick-shape networks.
- Simulation-driven data enrichment to improve machine-learning performance for robotics tasks.
- AI and compressed sensing for reconstructing pressure-sensor signals.
Platform portfolio
Project pages and code used for implementation
UAV Development
Design, integration, and testing of a multi-purpose UAV for surveillance, data acquisition, waypoint flight, and sensor payloads.
Open project OceanSim / UUV
A simulation and underwater robotics platform for control, survey missions, trajectory work, and multi-UUV experiments.
Open GitHub ORB-SLAM3
Visual and visual-inertial SLAM for real-time localization, map reconstruction, tracking, and robot navigation.
Open project LiDAR Perception
Laser-based ranging for 3D mapping, obstacle detection, and localization support in low-texture environments.
Open project Stereo Vision
Depth estimation from stereo cameras for navigation, obstacle awareness, geometric perception, and object detection.
Open project