INTER IIT
TECH MEET 14.0
PREPATHON PS
AERIAL
ROBOTICS
CM: Bhavesh Meghwal (8529797270)
PREPATHON 2025 | IIT(BHU)
PROBLEM STATEMENT
Introduction & Motivation
Modern UAV applications increasingly demand operation in GPS-denied environments
such as indoor facilities, urban canyons, underground structures, and areas with signal
jamming. Traditional GPS-based navigation fails in these scenarios, creating critical
operational gaps for surveillance, search-and-rescue, and mapping missions.
Vision-based navigation represents the next frontier in autonomous UAV technology,
enabling robust operation using onboard cameras and computer vision algorithms.
However, current implementations face significant challenges including scale ambiguity,
feature tracking failures, and integration with flight control systems. Typically, UAV
developers rely on two widely-used open-source flight controller stacks, ArduPilot and
PX4, to manage flight control, navigation, and sensor integration.
Recent developments in edge computing, AI-enhanced perception, and real-time
processing have created new opportunities for reliable vision-only navigation. The
ability to maintain stable flight, handle sensor failures gracefully, and provide accurate
positioning without external infrastructure is crucial for next-generation autonomous
systems.
As UAV operations expand into complex environments where GPS is unavailable or
unreliable, developing robust vision-based navigation becomes essential for mission
success and operational safety.
Problem Statement
Design and implement a complete vision-based autonomous navigation system capable
of GPS-denied flight operations with robust failsafe handling and recovery mechanisms.
Additionally, demonstrate control allocation modifications for motor failure scenarios in
ArduCopter flight stack.
Part 1 Vision-Based Navigation System in SITL Simulation 50%
of total points)
Vision Health Publisher Development
Implement MAVLink-based health monitoring that enables normal arming without force
commands.
Robust State Estimation
Develop position and velocity estimation algorithms that handle lighting variations,
motion blur, and feature tracking failures.
AERIAL ROBOTICS 1
EKF Integration
Integrate vision estimates with ArduPilot’s Extended Kalman Filter to maintain variance
thresholds below failsafe limits.
Part 2 PX4 Motor Failure Simulation 50% of total points)
Autonomous Takeoff
Achieve stable autonomous takeoff and hover at 20m altitude using PX4 flight stack.
Control Allocation Modification
Implement real-time motor mixing reconfiguration to disable one motor from the control
allocation matrix.
Flight Demonstration
Controlled fall after failure without failure detection or recovery algorithms.
PX4 Integration
Demonstrate understanding of PX4’s control allocation system and actuator output
pipeline.
Technical Requirements
Part 1 Vision-Based Navigation System
Phase 1 Vision Health Integration 25% of points)
• Implement proper VISION_POSITION_ESTIMATE and VISION_SPEED_ESTIMATE
publishers.
• Achieve normal arming without ”force” commands by providing stable health
metrics.
• Demonstrate stable EKF integration with vision source configuration.
• Provide real-time confidence metrics and quality assessment based on feature
count and tracking stability.
• Configure proper parameter files with GPS disabled and appropriate EKF3 source
settings.
• Maintain consistent 20–30Hz message frequency with proper NED frame
coordinate conversion.
2 AERIAL ROBOTICS
Phase 2 Autonomous Position Hold 25% of points)
• Maintain hover within 1.5m radius for 90 seconds using vision-only navigation in
Gazebo simulation.
• Handle EKF failsafe triggers gracefully with automatic recovery when variances
exceed FS_EKF_THRESH 0.8.
• Log synchronized vision/control data for performance analysis including EKF
variance monitoring.
• Implement scale-consistent velocity estimation using known altitude 10m takeoff
height).
• Achieve stable variance characteristics keeping compass, position, and velocity
variances below threshold.
Part 2 PX4 Motor Failure Implementation
Phase 1 PX4 Setup and Takeoff 15% of points)
• Configure PX4 firmware with standard quadrotor mixer configuration using Gazebo
SITL.
• Takeoff to 20m altitude with all four motors operational.
Phase 2 Motor Mixing Modification 35% of points)
• Implement runtime modification of PX4’s control allocation matrix to disable one
specific motor.
• Modify the control allocator module to remove motor from mixing.
Integration Requirements
• Both parts must use same simulation environment Gazebo with provided worlds).
• Code quality standards including proper documentation, modular design, and
parameter configuration.
• Real-time performance constraints: vision pipeline <60ms latency, control loops
maintain standard rates.
Submission Format
• A zip file containing 2 folders named PART_1 and PART_2.
• Each folder should contain:
– Proper documentation detailing deployment steps.
– Video demonstration.
AERIAL ROBOTICS 3
– README file and scripts.
Resources
• Basics of Quadcopter Controls by MATLAB
[Link]
• Ardupilot Dev Environment Setup in Linux:
[Link]
• PX4 Dev Environment Setup in Linux:
[Link]
• PX4 Control Architecture:
[Link]
4 AERIAL ROBOTICS