Opportunity brief
Event Description:
- Pixel2Path is an autonomous robotics competition where teams build a robot that turns digital images into physical drawings on paper, with no human intervention.
- It combines Machine Learning, Computer Vision, Path Planning, Robotics, and Mechatronics.
Online Submission: The teams have to submit the 'Documentation', 'BOM', and 'Video-Demonstration' of the bot. (Non-Eliminatory Round)
- Round 1: The robot receives an unseen image (PNG/NPY), processes it, plans an optimized path, and draws it on an A3 sheet within the allotted time.
- Round 2 (Draw): Each team gets an image with 40% of its strokes randomly removed. Their trained AI model predicts the missing strokes, and the robot draws the completed image (visible and predicted strokes) on A3 sheet.
- Round 2 (Predict): Teams get 10 unseen masked images and predict the originals. Each is scored by SSIM.
Teams get 100 image pairs two weeks before the event. On event day they get the full 5,000-pair dataset and a 60-minute window to train their model on Kaggle.
Rules:
- Teams must have 3 to 5 members, with one designated Team Lead for official communication.
- Teams must bring a fully assembled and tested robot (no on-day mechanical builds), a laptop with all software pre-installed, the trained model weights file, a 12V DC power supply, and a USB cable for Arduino.
- It must be a 3-axis plotter (X, Y, and a Z pen-lift) with independent control of each axis.
- The robot must have an Emergency Kill Switch that completely stops all motion.
- Training must be done in a Kaggle Notebook on event day with GPU acceleration, with Internet toggled OFF.
- Training must be from scratch, using only the provided 5,000-image dataset. Data augmentation is allowed.