Plastic and Paper Segregation Robot - LeRobot Global Hackathon 2025

This project, developed during the Hugging Face LeRobot Global Hackathon 2025 in London, secured the Technical Experts Award and 2nd place overall. As Team Episode-124 (named after our 124 collected data episodes), we tackled the critical challenge of automated waste segregation, specifically focusing on separating paper from mixed plastic-paper waste on a moving conveyor belt.

Key Achievements

• Won Technical Experts Award at LeRobot Global Hackathon 2025
• Secured 2nd Place in London competition
• Successfully implemented real-time waste segregation on moving conveyor
• Developed during an intensive 2-day deep-tech sprint

Technical Implementation

• Vision System: Implemented dual-camera setup (top and front views) for comprehensive object detection
• Robot Control: Utilized SO101 robot arm with fine-tuned GROOTN1.5 VLA model
• Dynamic Prediction: Developed algorithms to predict object positions during conveyor movement for precise grasping
• Data Collection: Created comprehensive dataset using leader-follower setup
  • 124 training episodes
  • Integrated camera feeds with robot joint angles
  • Real-time position tracking and prediction

Technical Evolution

• Initial Approach: Tested ACT model from scratch
• Intermediate Step: Experimented with fine-tuning smolVLA (faced prediction challenges)
• Final Solution: Successfully implemented fine-tuned GROOTN1.5 VLA model
  • Superior performance in dynamic object tracking
  • Accurate prediction of object positions during conveyor movement
  • Robust grasping strategy implementation

Technologies Used

• SO101 Robot Arm
• GROOTN1.5 VLA (Vision-Language-Action) model
• Dual Camera Vision System
• Leader-Follower Training Setup
• Advanced Motion Prediction Algorithms
• Real-time Control Systems

Project Resources

• Code Repository: LeRobot Implementation
• Dataset: Sort Boxes Conveyor Dataset v1

Team Members

• Om Kulkarni
• Nitheesh Kumar Senthilnathan
• Pranav Naik
• Prahalad Vijaykumar