Flipkart GRID 5.0 Robot
The software stack for a robotics finalist: YOLOv5 perception, RealSense depth, inverse kinematics, serial control, and QR-based sorting.
Flipkart GRID was the first project that made me feel the jump from writing code to building a system.
I was in my first year of college. The challenge sounded clean: detect an object, locate it in 3D space, pick it up, read the QR code, and place it in the right zone.
In practice, every step had to line up with hardware timing.
What I Built
I owned the software side of the robot:
- YOLOv5 detection on live camera input
- Intel RealSense depth for
(x, y, z)coordinates - inverse kinematics for robotic arm movement
- serial communication to Arduino-controlled hardware
- dual ESP32-CAM QR detection for sorting
- a state machine for the full pick-place cycle
The state machine mattered because the robot needed to behave like one system, not a pile of scripts.
What Was Hard
The difficult part was the handoff between layers.
Perception had uncertainty. Depth had noise. Inverse kinematics needed stable coordinates. Hardware commands had timing constraints. QR detection had to run in parallel with the rest of the workflow.
That project taught me early that systems work is mostly about the interfaces between parts.
Why It Stayed
We made the finals at IIT Madras, which was a big moment.
The project also showed me what kind of work I like: computer vision, physical hardware, and control logic all needing to align before anything useful happens.