Waste segregation remains inefficient, labor-intensive, and vulnerable to contamination. Manual sorting can be slow, expensive, and expose workers to hazardous environments, while inaccurate segregation reduces the purity and value of recyclable materials.
SmartSort AI addresses this problem by creating a closed-loop system that connects AI perception directly to physical robotic action.
The system uses a camera-based sensing pipeline to observe waste moving through a conveyor. RGB-D sensors capture high-resolution color and depth information, which is processed by Gemini's multimodal reasoning layer to classify the material. The AI then generates a structured sorting decision, maps the target coordinates, and sends the command to a robotic delta arm for precision interception.
The complete workflow consists of six stages:
Object Ingest — Waste enters through the automated conveyor. RGB-D Capture — Color and depth data are captured for analysis. AI Inference — SmartSort classifies the material. Target Mapping — The object's real-time coordinates are sent to the robotic arm. Precision Sort — Actuators divert the object into the appropriate recovery stream. Bin Validation — Weight and optical sensors verify placement and update inventory.
Unlike traditional rule-based sorting systems, SmartSort AI is designed around general-purpose visual reasoning, allowing the system to move beyond rigid, hard-coded object rules and adapt to different waste-stream conditions.
Our long-term vision is to evolve SmartSort from a hackathon prototype into a scalable smart-waste ecosystem for municipal authorities, industrial facilities, and recycling operations.
Inspiration
Traditional waste sorting relies heavily on manual labor and fixed automation rules. This creates three major problems: operational inefficiency, high labor dependency, and contamination of recyclable materials.
We wanted to explore a different approach:
What if a waste-sorting machine could actually understand what it sees instead of relying entirely on predefined rules?
That question led us to combine Gemini's multimodal reasoning with computer vision, real-time object localization, and robotic actuation.
Log in or sign up for Devpost to join the conversation.