Epoch8.co blog

ECOBOT: Recycling Robot with Computer Vision

ECOBOT: Recycling Robot with Computer Vision

ECOBOT employs an AI-driven robotic arm to automate waste sorting via computer vision.

How ECOBOT works

The ECOBOT system intakes a physical bag of trash, systematically opens it, and disperses its contents onto a conveyor. A strategically positioned robotic arm and camera setup detects and categorizes waste into classifications such as plastic, glass, metal, and cardboard. The robotic arm then mechanically separates the waste into dedicated containers based on its classification.

Project Objective

Develop and train an AI module to precisely recognize and categorize objects, followed by its integration into the robotic arm mechanism.

Technical Solution

The computer vision image recognition module has the capability to discern various objects, including specific items like glass bottles or crumpled paper. Once an object is identified, the system calculates the necessary trajectory for the robotic arm. Essentially, upon depositing trash, ECOBOT goes into action: it identifies, classifies, and instructs the robotic arm for appropriate action.

Challenges Addressed

Precision is paramount. The neural network is fine-tuned to determine object coordinates based on image pixels. The core task goes beyond mere identification: the computer vision module translates visual data into actionable commands for the robotic arm. For instance, post object detection, it recalculates coordinates and issues sequential instructions, such as angular adjustments, vertical movements, or specific grips.

Operational Workflow

  1. Video feed acquisition from the camera.
  2. Object recognition involving detection and classification.
  3. Coordinate recalibration, converting image pixel data into robotic control signals.
  4. Sequence integration, where all steps are translated into mechanical commands for the robot – a collaboration with our robotics specialists.

Capabilities & Outcomes

The ECOBOT system:

  • Processes 4 frames per second.
  • Identifies 8 distinct object types.

Implemented Technologies

  • CenterNet
  • ResNet50
  • Jetson Nano
  • Robot Operating System (ROS)
Case Studies