Final project demonstration

CV autonomous laps:

GPS laps tuning:

Project Overview

Timeline: January 2025 - February 27, 2026 (ongoing)

RoboCar is an autonomous self-driving robot car project built on the Donkeycar framework. The platform was developed in two parallel tracks: computer-vision-based autonomous laps and GPS-guided laps. The work focused on practical autonomy integration, hardware/software interfacing, and robust tuning in real track conditions.

Technologies & Skills Demonstrated

  • Autonomous Driving Frameworks
    • Built and tuned pipelines with the Donkeycar stack for real-time autonomous control.
    • Configured CV-based inference loop for lap execution.
  • Computer Vision and ML Deployment
    • Collected and prepared driving datasets for model training/inference.
    • Tuned autonomous behavior for repeatable CV-guided laps.
  • GPS Navigation and Control
    • Implemented and tuned GPS-guided lap routines.
    • Evaluated trajectory consistency and refined control parameters iteratively.
  • Embedded and Radio Integration
    • Configured an Arduino bridge to expose ELRS radio controls as joystick inputs to the Raspberry Pi.
    • Verified end-to-end control handoff between manual and autonomous operation modes.

Project Development

CV autonomous lap stack

The CV branch focused on closed-loop autonomous driving from onboard vision, with iterative data collection, model updates, and behavior refinement for smoother laps.

GPS-guided lap stack

The GPS branch focused on repeatable lap behavior under waypoint/position guidance and continuous tuning of control parameters.

RC-to-compute bridge integration

A custom Arduino bridge was configured so ELRS radio input appears as a joystick interface on the Raspberry Pi, enabling practical control-mode switching and safer testing workflows.

Challenges + Solutions

  • Challenge: Stability and consistency in autonomous laps
    • Solution: repeated track testing and parameter tuning across both CV and GPS pipelines.
  • Challenge: Hardware/software interfacing for remote control bridge
    • Solution: custom Arduino bridging and interface validation so ELRS data was consistently exposed to the host compute stack.
  • Challenge: Keeping development moving across two autonomy approaches
    • Solution: maintained separate repositories and workflows for GPS and CV modes while sharing test learnings.

Future Improvements

  • Add detailed telemetry logging and replay tooling for faster root-cause analysis.
  • Integrate stronger failure recovery behavior for edge cases during autonomous runs.
  • Publish a unified documentation repo with hardware wiring, configs, and tuning recipes.