Lunabotics Navigation Package Team May14-20 Advisor: Dr. Koray Celik Clients: ISU Lunabotics Club, Vermeer Company
Project Scope NASA Lunabotics Mining Competition For university-level students to design and build a mining robot that can traverse the simulated Martian chaotic terrain Points are awarded for autonomy Landmarks allowed for navigation
Project Scope Our Clients Lunabotics Club Vermeer Company o Large industrial and agricultural equipment
Project Scope ISU Lunabotics Club Bump detectors Laser detector for orientation No obstacle detection LabView
Project Scope Our Solution Create prototype navigation system using computer vision techniques for navigation and object detection Microsoft Kinect for prototype hardware solution Open-source software
Design Platform Laptop Computer o 2.26 GHz Dual Core Processor o 4 GB RAM Microsoft Kinect o RGB Camera o Infrared Optics Robot ‘Virgil’
Design Virgil Built by Dr. Koray Celik Representative of Lunabotics robot Modified Kinect Motor Controller o 4 Motors Encoder Controller o 6 Encoders Li-Ion Battery Powered
Design Software Ubuntu Linux LTS OpenCV OpenNI PropulsionClass
Design
Landmark Detection Acquisition Module Color Threshold Canny Convolution Operator Boundary Tracing Rectilinear Object Recognition
Landmark Detection CannyThreshold
Landmark Detection Contour All Poly Filter
Landmark Detection ●Unobtrusive landmark, not likely to be found naturally ●Searching for rectilinear shapes with overlapping midpoints ●State-space reduction to reduce noise
Design Obstacle Detection Acquire depth map Smoothing Regression Least Squares Plane Fitting
Obstacle Detection Plane Fitting Least Squares Regression Smoothed Depth Mat Normal Floor Plane Blue - Object Orange - Floor
Obstacle Detection Threshold regression values Contour Filter Area of bounding rectangle
Path Generation Finite State Machine Landmark must be in sight upon start Drives towards landmark, avoids obstacles that appear in its path Maintains spacial awareness to landmark in case it loses sight
Path Gen 1.Find landmark 2.Move towards the landmark 3.Obstacle detected 4.Proceed on alternate path 5.Turn back to relocate landmark 6.Head to landmark
Accomplishments Core Requirements Robot Movement o Smooth Acceleration and turning o Position and heading Landmark detection o Recognize and distinguish between our two landmarks Obstacle detection o Recognise obstacles above 4 cm tall o Recognize Landmarks as not obstacles Navigation o Obstacle avoidance
Challenges Landmark Detection o Noise reduction Obstacle Detection o Uneven terrain issues Encoder o Set back by hardware (group is more software oriented OpenNI Bug o Appears around 3 minute mark o Prevents code from scalability in terms of computational complexity o Caused by XnSensorServer
Testing Simulated Arena 3 m X 2 m area Virgil is representative of Lunabotics Club’s Robot Method Test movement Test landmark detection, eliminate false-positives Test object detection Test obstacle avoidance
Demonstration
What’s Next? Introduce Improved Hardware o 2.5-D Lidar o High Definition RGB Camera Integrate Encoder Support Improve Navigation Algorithm
Questions?