We are thrilled to announce that TensorForest has won the WRO (World Robot Olympiad) Toronto Regional Competition in the Future Innovators category! 🎉 Our team's hard work, dedication, and innovative approach have been recognized as we now qualify for the national level. We are excited to continue refining our project and showcasing the potential of combining drone technology and machine learning to address environmental challenges.
TensorForest is a Python-based application developed by Team #154 (Arnnav Kudale, Shayaan Azeem, and Rohanth Marem) designed to address the increasing threat of wildfires using automated drone technology and machine learning. Originating from a shared passion for robotics and environmental sustainability, the project aims to provide accurate and accessible data for wildfire prevention.
The development of TensorForest faced several challenges:
- Hardware Integration: Compatibility issues and precise calibration of flight controllers, GPS modules, and camera systems posed significant challenges.
- Software Development: The development of machine learning models, flight planning algorithms, and image processing scripts required thorough research and debugging.
- Field Testing: Logistical challenges and safety considerations arose during real-world testing in forest environments.
The team adopted a systematic approach to overcome these challenges:
- Research and Planning: Extensive research informed development decisions regarding technology selection and wildfire prevention methods.
- Collaborative Work: Tasks were divided based on expertise, ensuring comprehensive coverage of hardware, software, and machine learning components.
- Iterative Development: The project evolved through continuous improvement based on feedback from testing and evaluation.
- Adaptability: The team remained flexible and open to adjustments, facilitating the resolution of unforeseen challenges and the incorporation of new insights.
The project was developed using a combination of hardware construction, software development, machine learning, and field testing:
- Hardware Construction: The drone was assembled using a prebuilt carbon fiber kit and selected components for efficiency, precision, and compatibility.
- Software Development: Open-source frameworks like ArduPilot and TensorFlow Lite were utilized for autonomous flight control and object detection.
- Machine Learning: Custom models were trained using TensorFlow and Coral Edge TPU to classify vegetation types and assess flammability levels.
- Field Testing: Tests were conducted to validate the drone's performance in capturing imagery and generating flammability maps.
The team plans to further enhance and expand the project:
- Enhanced Sensor Integration: Exploring the use of LiDAR or multispectral sensors to improve data accuracy and expand capabilities.
- Optimization and Scaling: Refining algorithms for faster processing and scalability to handle larger datasets and diverse regions.
- Community Engagement: Collaborating with authorities and organizations to deploy TensorForest in wildfire-prone areas and gather feedback.
- Commercialization: Exploring partnerships to promote widespread adoption for wildfire management.
To get a local copy up and running, follow these simple steps:
- Clone the repo
git clone https://github.com/ultratrikx/tensorforest.git - Install Python packages
pip install -r requirements.txt - Run the application
python main.py
For a more detailed explanation of our project, including our methodologies, challenges faced, and future steps, please refer to our Project Report.
- Rohanth Marem - @rohanthmarem
- Shayaan Azeem - @shayaan_azeem
- Arnnav Kudale - @blazecoding2009
Project Link: https://github.com/ultratrikx/tensorforest