Case Study // autonomous-drone

Vyoma

Custom ROS 2 drone autopilot with YOLOv8 semantic vision and context-aware navigation on NVIDIA Jetson Nano.

AutopilotPX4

VisionYOLOv8

RuntimeJetson Nano

SimGazebo

DEMO_RECORDING

[ VIDEO ASSET REQUIRED ]

02 // Architecture Execution

Built a custom ROS 2 autopilot stack on NVIDIA Jetson Nano, integrating YOLOv8 for edge-based semantic object detection. The system fuses visual context with PX4 flight control to enable intelligent, scene-aware navigation — tested end-to-end in Gazebo simulations.

01 // The Problem Context

Standard UAV autopilots rely purely on pre-programmed waypoints and LIDAR — they cannot semantically understand their environment or make context-aware decisions in real-time.

System Stack

ROS 2YOLOv8PX4GazeboPythonNVIDIA Jetson

core_module.ts

# ROS 2 YOLOv8 Semantic Detection Node
class SemanticVisionNode(Node):
    def __init__(self):
        super().__init__('vyoma_vision')
        self.model = YOLO('yolov8n.pt')  # Edge-optimized
        self.sub = self.create_subscription(
            Image, '/camera/image_raw', self.detect_cb, 10)

    def detect_cb(self, msg):
        results = self.model(bridge.imgmsg_to_cv2(msg))
        self.publish_context(results)  # Feed to PX4 autopilot

Source Code