Inspiration
When tackling the John Deere Mechathon track, we looked to the core principles of modern agricultural and construction machinery: precision, data-driven operation, and uncompromising safety. We wanted to bring that industrial spirit to a micro-scale. TerraGlide was inspired by the idea that even a sandbox-sized earthmover needs smart reflexes. We set out to build a machine that doesn't just push dirt around, but actively monitors its environment to keep humans safe and prevent itself from getting bogged down in tough terrain.
What it does
TerraGlide is a smart, semi-autonomous sand-leveling rover. Operators can seamlessly pilot the machine using a custom-built UI/UX accessible via laptop or smartphone.
While under teleoperation, the robot runs continuous background safety and operational checks:
- Pedestrian Safety Protocol: Using an onboard webcam and YOLO object detection, TerraGlide monitors its perimeter. If a person gets too close to the working machine, it triggers high-visibility LED flashes to alert bystanders to clear the active zone.
- Autonomous Terrain Mitigation: Leveling sand is unpredictable. We integrated an ultrasonic sensor array that reads the terrain immediately in front of the rover. If the sensors detect that the front dozer is digging too deep and risks stalling the motors, the system overrides manual control to autonomously lift the blade, preventing the machine from getting stuck.
How we built it
Our foundation was the SunFounder Galaxy-RVR kit, which provided the base chassis, drive motors, and initial sensor suite (webcam, ultrasonic). To bring TerraGlide to life, we heavily modified the base kit:
- Compute & Control: We utilized both an ESP32 and an Arduino to handle Wi-Fi communication, sensor polling, and motor actuation, linking our custom control UI to the hardware.
- AI Vision Stack: The camera feed is processed through a YOLO model to handle the real-time pedestrian detection for our safety alerts.
- Custom Fabrication: The base kit wasn't built for earthmoving. We had to design our own custom front dozer and rear grader attachments. Since we didn't have existing CAD files for the SunFounder chassis, we manually measured every mounting point and dimension using calipers, designed the attachments in CAD, and rapid-prototyped them using laser cutting and 3D printing.
Challenges we ran into
Integrating complex software with physical, messy mechanics in a hackathon timeframe threw several curveballs at us:
- The Sand Trap: When we first attached our custom rear grader, the drag was immense. The sand physics caused the rover to dig itself into a hole and get completely stuck, forcing us to rethink our weight distribution and blade depth.
- Protocol Puzzles: We spent a massive amount of time troubleshooting and reverse-engineering the proprietary communication protocol the SunFounder kit uses to send commands to the rover so we could bypass it for our custom UI.
- Blind Fabrication: Designing mechanical attachments from scratch without reference CAD models meant our manual measurements had to be flawless to ensure the 3D printed and laser-cut parts fit the chassis on the first assembly.
Accomplishments that we're proud of
We are incredibly proud of bridging the gap between high-level AI and low-level hardware control. Successfully running a YOLO vision model alongside real-time ultrasonic hardware interrupts—while managing an ESP32 Wi-Fi control stack—is a huge win. We're also thrilled that our manually measured CAD designs mated perfectly with the base rover, transforming a standard kit into a legitimate, functional piece of micro-heavy-machinery.
What we learned
Building TerraGlide gave us massive respect for the engineering at John Deere. We learned firsthand that hardware-software integration is notoriously tricky, especially when dealing with the unpredictable physical resistance of sand. We also learned how critical localized, data-driven fail-safes (like our auto-blade-lift) are to keeping machinery operational in the field.
What's next for TerraGlide
In the future, we want to upgrade TerraGlide from semi-autonomous to fully autonomous. We plan to implement pathfinding algorithms so the rover can autonomously map a sandbox and execute a complete leveling grid. We'd also love to upgrade the chassis with stronger motors and custom treads to handle even more aggressive grading tasks.
Log in or sign up for Devpost to join the conversation.