Project Overview
A 1:10 scale autonomous vehicle prototype powered by NVIDIA Jetson Nano running TensorRT-optimized YOLOv8 computer vision and C++ PID motor control.
Problem Statement & Engineering Motivation
Developing an affordable, low-latency autonomous vehicle testbed that can detect track boundaries and obstacles in real-time on edge compute without relying on expensive cloud servers.
Technical Architecture & Implementation
Built on an RC chassis with an NVIDIA Jetson Nano. Camera feeds are processed through an optimized YOLOv8 neural network for obstacle classification, while a custom C++ feedback controller handles steering actuation and PWM speed regulation.
Key Technical Features
- Real-time object detection and obstacle tracking on edge hardware
- Adaptive computer vision lane-following algorithms
- Dynamic obstacle avoidance with emergency braking thresholds
- Serial telemetry logger for velocity and inference metrics
Technology & Component Stack
- NVIDIA Jetson Nano
- Python 3.10
- OpenCV
- TensorRT
- C++
- PWM Controllers
- LiPo Battery Management
Outcomes & Engineering Lessons Learned
- Low-cost 2D LiDAR often suffers from surface absorption on dark objects, making camera vision essential for redundancy.
- Quantizing neural models to FP16 via TensorRT significantly improves inference throughput on embedded Jetson Maxwell GPUs.
Related Research Study
This engineering build is connected with academic research investigation: /research/autonomous-vehicles-qatar