GPS-free Autonomous Drone Navigation

Introduction

Project Goal

The goal of our project is to develop a GPS-less autonomous navigation system for a Tello drone that uses real-time camera input for environmental awareness and object tracking. The system enables the drone to visually detect a target object, navigate toward it, and land precisely when it reaches a predefined proximity threshold—all without relying on GPS. This involves implementing robust sensing, planning, and actuation components to ensure the drone can perform in GPS-denied environments such as indoors or in constrained outdoor areas.

Interesting Challenges

• Navigating in GPS-Denied Environments: Most autonomous drones rely heavily on GPS for localization and navigation. Designing a system that operates without GPS requires innovation in computer vision and state estimation.
• Object Detection and Tracking in Real-Time: The system must process video frames in real-time, perform noise reduction, and track the target object despite changing lighting conditions, motion blur, and dynamic backgrounds.
• Combining SLAM and Object Tracking: Balancing the integration of SLAM (Simultaneous Localization and Mapping) for navigation and object tracking for interaction poses an interesting problem in resource management and system design.
• Precise Landing: Landing safely on or near the target using only visual feedback is a non-trivial problem that involves accurate distance estimation, trajectory planning, and control. These challenges push the boundaries of what a lightweight and affordable drone platform like the Tello can achieve, making the project a valuable exploration into advanced robotics.

Real-World Applications

The work from our project has numerous real-world applications, particularly in environments where GPS signals are unavailable or unreliable. In search and rescue operations, GPS-less drones can navigate through disaster-stricken areas such as collapsed buildings or dense forests to locate survivors, equipment, or specific markers. Similarly, in indoor navigation, drones equipped with visual navigation systems can be used for warehouse management, inventory tracking, or package delivery in large facilities or urban environments where GPS access is limited. In the field of agriculture, GPS-free drones can autonomously navigate greenhouses or tightly planted crop fields to monitor plant health, identify issues, or perform precise, localized spraying. Additionally, in surveillance and inspection, such drones can operate in hard-to-reach or confined spaces, such as industrial machinery, power plants, or other areas where GPS signals are obstructed, providing a valuable tool for routine maintenance and monitoring.

Design

Functionality and Design Choices

Our desired function for the drone is to enable it to detect certain objects, fly towards them, and perform specific actions. Since the Tello drone is small and adding additional hardware might compromise its robustness, we decided that the action would be landing in front of the target.

Design Limitations

Implementation

Hardware Specifications

Algorithms Used

1. Camera Streaming Algorithm

Purpose: Establishes a real-time video stream from the Tello drone for visual tracking.

Steps:

• Connection Setup: Connect to Tello drone and start video stream
• Stream Handling: Manage video stream with av library
• Frame Retrieval: Extract and convert frames to OpenCV format
• Error Management: Handle AVError exceptions gracefully

2. Frame Processing Algorithm

Purpose: Process frames for color-based target detection and tracking.

Steps:

• Pre-Processing: Apply blur and convert to HSV
• Target Detection: Use HSV bounds and morphological operations
• Contour Analysis: Find and analyze target contours
• Offset Calculation: Determine target position relative to center
• Visualization: Draw tracking indicators and annotations

3. Drone Exploration Algorithm

Purpose: Enable autonomous target search when not in view.

Steps:

• Initialization: Begin spinning motion and altitude maintenance
• Detection Transition: Monitor frames and switch to tracking
• Fallback: Return to exploration if target lost
• Adaptive Movement: Adjust search patterns dynamically

4. PID Control Algorithm

Purpose: Ensure precise movement control based on target position.

PID Values:

• Horizontal Control: Kp = -0.5, Ki = 0.01
• Vertical Control: Kp = 0.001, Ki = 0.0001

Steps:

• Error Calculation: Compute and normalize positional error
• PID Components: Calculate P, I, and D terms
• Output Computation: Generate controlled movement commands
• Command Execution: Apply yaw and throttle adjustments
• Dynamic Updates: Maintain real-time error correction

System Workflow

Step 1: Initialization

• The drone connects to the Tello API, sets up its video stream, and initializes parameters for exploration and tracking.
• The system initializes the PID controllers for yaw (horizontal rotation) and throttle (vertical movement).

Step 2: Cup Finding State

• The drone begins in the Cup Finding State, spinning in place to search for the target (the cup).
• The camera stream is processed frame by frame and the frames are analyzed for the presence of a red-colored object using HSV filtering.
• If no target is detected, the drone continues spinning and stabilizing its altitude using a low throttle setting.

Step 3: Target Detection

• When the target is identified, the system transitions to the Tracking State.
• The drone stops spinning and begins using the PID controller to align with the target's position.

Step 4: Tracking State

• In the Tracking State, the drone actively follows the target using PID control.
• The frame's offset data (x, y) and the target's size (radius) are fed into the PID controllers.
• The PID outputs adjust the drone’s yaw and throttle to keep the target centered in the frame.

Step 5: Proximity Detection

• The system monitors the size of the detected target (percentage of the screen area).
• When the target occupies a predefined area of the frame (e.g., 60%), the drone initiates a landing sequence.
• If the target is lost during tracking, the system reverts to the Cup Finding State.

Step 6: Landing

• Once the target is deemed close enough, the drone lands autonomously by cutting forward motion and reducing throttle to zero.

States of Operation

Cup Finding State

Objective: Search for and identify the target cup.

Actions:

• Clockwise spinning search
• Altitude maintenance
• Continuous frame processing
• HSV color filtering for detection

Transition Condition: Target detected in frame

Tracking State

Objective: Lock onto target and align position.

Actions:

• PID controller adjusts yaw based on horizontal offset
• Throttle adjustment based on vertical offset
• Forward movement when target below threshold
• Continuous target tracking

Transition Condition: Target size >60% or target lost

Landing State

Objective: Safe landing when target is close.

Actions:

• Forward velocity zeroing
• Gradual throttle reduction
• Ground detection landing

Transition Condition: Ground contact detected

PID Control System

Implementation of dual-axis PID control:

• Horizontal Control: Kp = -0.5, Ki = 0.01
• Vertical Control: Kp = 0.001, Ki = 0.0001
• Target size-based distance control
• Automatic mode switching for target loss