Develop a C++ Autonomous Drone Tracking System (PX4 + TensorRT + Gazebo Simulation)
Budget: €1,500 – €3,000 EUR
Project Overview
We are looking for an experienced C++ developer with strong expertise in autonomous drone systems, PX4 Offboard control, computer vision, and sensor fusion to join the Lumetrix project.
Lumetrix is an advanced real-time autonomous tracking and interception system designed for drones running PX4 and NVIDIA Jetson Orin NX. The system will detect, track, and intercept a target drone using data from multiple sensors, including depth cameras, lateral cameras, and a 4D radar.
All development, collaboration, and validation will take place directly through our GitHub repository. Progress will be verified through commits, code reviews, and milestone-based demonstrations.
The official project start will occur once the related research grant or private sponsorship is confirmed. Candidate selection and technical evaluation will take place beforehand.
Hardware and Sensors
- Companion Computer: NVIDIA Jetson Orin NX
- Flight Controller: PX4 on Holybro Kakute H7
- Main Camera: Intel RealSense D435i (RGB + Depth + IMU)
- Lateral Cameras: Two IMX219 (120° FOV)
- Radar: TI IWR6843 4D mmWave Radar
- Communication: MAVLink over UART/UDP using MAVSDK-C++
Software Stack
- Language: C++17/20
- Frameworks and Libraries: TensorRT, OpenCV, Eigen3, MAVSDK-C++, spdlog, yaml-cpp
- Simulation: PX4 SITL + Gazebo Garden (dual-drone simulation environment)
- Build System: CMake
- Repository: GitHub (mandatory for collaboration and review)
Main Development Objectives
Object Detection
- Implement YOLOv10 inference using the TensorRT C++ API.
- Handle pre-processing and post-processing, including resizing and non-max suppression.
- Support both real camera input and simulated Gazebo video feed.
Tracking and Prediction
- Implement ByteTrack or StrongSORT in C++.
- Integrate an Extended Kalman Filter for position and velocity estimation.
- Handle temporary occlusions and target recovery.
PX4 Offboard Control
- Integrate MAVSDK-C++.
- Implement Follow and Intercept flight modes.
- Add automatic failsafe and safety modes (Loiter, Return-to-Launch).
Sensor Fusion
- Combine data from the Intel D435i, two IMX219 cameras, and a 4D radar.
- Synchronize timestamps and run a multi-sensor Extended Kalman Filter.
- Output a fused 3D target pose in NED coordinates.
Dual-Drone Simulation
- Simulate both a tracker drone and a target drone in Gazebo.
- Implement trajectory generation and testing scripts.
- Validate end-to-end system performance in simulation.
Code Quality and Documentation
- Deliver clean, modular, and well-commented C++ code.
- Use YAML-based configuration files and spdlog for logging.
- All commits and pull requests must be handled via GitHub.
Deliverables and Milestones
M1 – Model Setup and TensorRT Integration (500 €)
Objective: Prepare the YOLOv10 model for real-time inference on NVIDIA Jetson Orin NX.
Tasks:
- Select or obtain pretrained YOLOv10 weights (COCO or similar dataset).
- Convert the model from PyTorch → ONNX → TensorRT .engine optimized for Orin NX (FP16 preferred).
- Integrate the TensorRT engine into a standalone C++ inference module using OpenCV for image capture.
- Run an inference test on a sample video or camera stream, logging FPS and latency results.
Deliverables:
- Working C++ inference code (CMake build system).
- TensorRT .engine file and conversion scripts.
- Short test log or video showing correct detections.
- All commits pushed and verified on GitHub.
Payment: 500 € upon successful demonstration and GitHub code review.
M2 – Tracking and EKF Prediction (400 €)
Objective: Implement target tracking and predictive filtering.
Tasks:
- Implement a tracking algorithm (ByteTrack or StrongSORT) in C++.
- Integrate an Extended Kalman Filter (EKF) to estimate target position and velocity.
- Ensure robustness to temporary occlusions or target loss.
- Output position, velocity, and ID of tracked objects.
Deliverables:
- Functional tracking module with real-time output.
- Log files showing consistent tracking of a single target.
- Source code properly commented and pushed to GitHub.
Payment: 400 € upon successful code validation and review.
M3 – PX4 Offboard Control Integration (500 €)
Objective: Enable autonomous flight control using MAVSDK-C++ with PX4.
Tasks:
- Implement PX4 Offboard interface via MAVSDK-C++.
- Subscribe to target position data from the tracking module.
- Publish velocity or position setpoints to PX4.
- Add automatic safety behaviors (Loiter and RTL modes).
Deliverables:
- Working MAVSDK control module.
- Drone follows target in PX4 SITL simulation.
- Source code committed and validated on GitHub.
Payment: 500 € upon successful flight test in simulation.
M4 – Dual-Drone Gazebo Simulation (500 €)
Objective: Develop a full simulation environment with both tracker and target drones.
Tasks:
- Create a Gazebo world featuring two PX4 SITL drones (tracker and target).
- Implement launch scripts for initializing both instances.
- Add a Python or C++ trajectory generator for the target drone.
- Integrate simulated camera and telemetry streams for testing.
Deliverables:
- Functional dual-drone Gazebo environment.
- Documentation on how to start and control the simulation.
- Code and assets committed to GitHub repository.
Payment: 500 € upon successful verification of simulation behavior.
M5 – Multi-Sensor Fusion (Radar + Cameras) (700 €)
Objective: Implement robust fusion of camera and radar data for reliable tracking.
Tasks:
- Fuse Intel RealSense D435i depth data, two IMX219 lateral camera feeds, and TI IWR6843 4D radar inputs.
- Perform time synchronization and spatial calibration among all sensors.
- Extend the EKF to integrate multi-sensor data.
- Test system performance under partial occlusion or high-speed motion.
Deliverables:
- Operational fusion module producing stable 3D target positions.
- Log files and accuracy validation results.
- Fully commented source code committed to GitHub.
Payment: 700 € upon code validation and performance review.
M6 – Documentation and Final Optimization (400 €)
Objective: Finalize the project for deployment and ensure maintainability.
Tasks:
- Review and clean all source code.
- Write clear technical documentation and README files.
- Verify build process (CMake) and configuration files (YAML).
- Optimize performance and eliminate redundant dependencies.
Deliverables:
- Complete and documented codebase ready for deployment.
- System passes final functional verification.
- All work verified and merged through GitHub.
Payment: 400 € upon delivery and code approval.
Total Project Budget
3000 € – 3500 €
All payments are milestone-based and released only after successful code review, compilation, and demonstration via GitHub commits.
We are looking for an experienced C++ developer with strong expertise in autonomous drone systems, PX4 Offboard control, computer vision, and sensor fusion to join the Lumetrix project.
Lumetrix is an advanced real-time autonomous tracking and interception system designed for drones running PX4 and NVIDIA Jetson Orin NX. The system will detect, track, and intercept a target drone using data from multiple sensors, including depth cameras, lateral cameras, and a 4D radar.
All development, collaboration, and validation will take place directly through our GitHub repository. Progress will be verified through commits, code reviews, and milestone-based demonstrations.
The official project start will occur once the related research grant or private sponsorship is confirmed. Candidate selection and technical evaluation will take place beforehand.
Hardware and Sensors
- Companion Computer: NVIDIA Jetson Orin NX
- Flight Controller: PX4 on Holybro Kakute H7
- Main Camera: Intel RealSense D435i (RGB + Depth + IMU)
- Lateral Cameras: Two IMX219 (120° FOV)
- Radar: TI IWR6843 4D mmWave Radar
- Communication: MAVLink over UART/UDP using MAVSDK-C++
Software Stack
- Language: C++17/20
- Frameworks and Libraries: TensorRT, OpenCV, Eigen3, MAVSDK-C++, spdlog, yaml-cpp
- Simulation: PX4 SITL + Gazebo Garden (dual-drone simulation environment)
- Build System: CMake
- Repository: GitHub (mandatory for collaboration and review)
Main Development Objectives
Object Detection
- Implement YOLOv10 inference using the TensorRT C++ API.
- Handle pre-processing and post-processing, including resizing and non-max suppression.
- Support both real camera input and simulated Gazebo video feed.
Tracking and Prediction
- Implement ByteTrack or StrongSORT in C++.
- Integrate an Extended Kalman Filter for position and velocity estimation.
- Handle temporary occlusions and target recovery.
PX4 Offboard Control
- Integrate MAVSDK-C++.
- Implement Follow and Intercept flight modes.
- Add automatic failsafe and safety modes (Loiter, Return-to-Launch).
Sensor Fusion
- Combine data from the Intel D435i, two IMX219 cameras, and a 4D radar.
- Synchronize timestamps and run a multi-sensor Extended Kalman Filter.
- Output a fused 3D target pose in NED coordinates.
Dual-Drone Simulation
- Simulate both a tracker drone and a target drone in Gazebo.
- Implement trajectory generation and testing scripts.
- Validate end-to-end system performance in simulation.
Code Quality and Documentation
- Deliver clean, modular, and well-commented C++ code.
- Use YAML-based configuration files and spdlog for logging.
- All commits and pull requests must be handled via GitHub.
Deliverables and Milestones
M1 – Model Setup and TensorRT Integration (500 €)
Objective: Prepare the YOLOv10 model for real-time inference on NVIDIA Jetson Orin NX.
Tasks:
- Select or obtain pretrained YOLOv10 weights (COCO or similar dataset).
- Convert the model from PyTorch → ONNX → TensorRT .engine optimized for Orin NX (FP16 preferred).
- Integrate the TensorRT engine into a standalone C++ inference module using OpenCV for image capture.
- Run an inference test on a sample video or camera stream, logging FPS and latency results.
Deliverables:
- Working C++ inference code (CMake build system).
- TensorRT .engine file and conversion scripts.
- Short test log or video showing correct detections.
- All commits pushed and verified on GitHub.
Payment: 500 € upon successful demonstration and GitHub code review.
M2 – Tracking and EKF Prediction (400 €)
Objective: Implement target tracking and predictive filtering.
Tasks:
- Implement a tracking algorithm (ByteTrack or StrongSORT) in C++.
- Integrate an Extended Kalman Filter (EKF) to estimate target position and velocity.
- Ensure robustness to temporary occlusions or target loss.
- Output position, velocity, and ID of tracked objects.
Deliverables:
- Functional tracking module with real-time output.
- Log files showing consistent tracking of a single target.
- Source code properly commented and pushed to GitHub.
Payment: 400 € upon successful code validation and review.
M3 – PX4 Offboard Control Integration (500 €)
Objective: Enable autonomous flight control using MAVSDK-C++ with PX4.
Tasks:
- Implement PX4 Offboard interface via MAVSDK-C++.
- Subscribe to target position data from the tracking module.
- Publish velocity or position setpoints to PX4.
- Add automatic safety behaviors (Loiter and RTL modes).
Deliverables:
- Working MAVSDK control module.
- Drone follows target in PX4 SITL simulation.
- Source code committed and validated on GitHub.
Payment: 500 € upon successful flight test in simulation.
M4 – Dual-Drone Gazebo Simulation (500 €)
Objective: Develop a full simulation environment with both tracker and target drones.
Tasks:
- Create a Gazebo world featuring two PX4 SITL drones (tracker and target).
- Implement launch scripts for initializing both instances.
- Add a Python or C++ trajectory generator for the target drone.
- Integrate simulated camera and telemetry streams for testing.
Deliverables:
- Functional dual-drone Gazebo environment.
- Documentation on how to start and control the simulation.
- Code and assets committed to GitHub repository.
Payment: 500 € upon successful verification of simulation behavior.
M5 – Multi-Sensor Fusion (Radar + Cameras) (700 €)
Objective: Implement robust fusion of camera and radar data for reliable tracking.
Tasks:
- Fuse Intel RealSense D435i depth data, two IMX219 lateral camera feeds, and TI IWR6843 4D radar inputs.
- Perform time synchronization and spatial calibration among all sensors.
- Extend the EKF to integrate multi-sensor data.
- Test system performance under partial occlusion or high-speed motion.
Deliverables:
- Operational fusion module producing stable 3D target positions.
- Log files and accuracy validation results.
- Fully commented source code committed to GitHub.
Payment: 700 € upon code validation and performance review.
M6 – Documentation and Final Optimization (400 €)
Objective: Finalize the project for deployment and ensure maintainability.
Tasks:
- Review and clean all source code.
- Write clear technical documentation and README files.
- Verify build process (CMake) and configuration files (YAML).
- Optimize performance and eliminate redundant dependencies.
Deliverables:
- Complete and documented codebase ready for deployment.
- System passes final functional verification.
- All work verified and merged through GitHub.
Payment: 400 € upon delivery and code approval.
Total Project Budget
3000 € – 3500 €
All payments are milestone-based and released only after successful code review, compilation, and demonstration via GitHub commits.