Robotics: Vision-Aided Navigation and Motion Path Planning on Low-End Android Hardware
Download as PPTX, PDF1 like283 views
This presentation provides an overview of ACME Robotics' 2017-2018 robotics team. It discusses their involvement in the FIRST and FTC robotics competitions. It then details how the team uses computer vision with OpenCV to implement tasks like jewel and cryptobox detection. It explains concepts like image representation, thresholding, color spaces, contours and morphological operations. Finally, it discusses the team's use of motion profiling and PID control to implement accurate motion control for their robot subsystems.
Robotics: Vision-Aided Navigation and Motion Path Planning on Low-End Android Hardware
- 1. ACME Robotics We do robot stuff A Presentation for Nevada County Tech Connection
- 2. 2017 - 2018 Team
- 3. FIRST: For Inspiration and Recognition of Science and Technology
- 4. FTC: First Tech Challenge
- 8. What is Computer Vision? ● “Gaining high-level understanding from digital images or videos” ● Includes video tracking, object recognition, 3D pose estimation, motion estimation, and image restoration. ● We’re mainly concerned with object recognition Acquire ActThink Computer Vision
- 9. Motivation ● Robot localization (estimating the robot’s position on the field) is very useful in autonomous ○ Dead reckoning approaches are only accurate for short movements ○ Other advanced sensing techniques are not feasible (e.g., lidar is illegal) ● Major field elements are composed of simple shapes with solid, saturated colors ○ This makes simple image processing feasible (i.e., no CNNs needed)
- 10. Image Representation ● Computers use matrices to represent image data ● Grayscale images are stored as a 2D matrix with each cell storing the intensity of the corresponding pixel ● Color images have multiple channels (e.g., RGB) with a separate matrix for each channel ● For example, the red channel matrix stores the intensity of red in each pixel 0 255 8-bit gray palette RGB channels
- 11. OpenCV ● An open source computer vision (CV) and machine learning software library ● Very common in industry ● Provides implementations of many CV algorithms and supporting data structures ● These operations can be assembled and composed together to form pipelines
- 12. Basic Example: Jewel Vision ● Jewel detection is a useful task in this year’s game ● Although a color sensor can be used, they are somewhat unreliable and difficult to position ● Since the position of the jewels is fixed, we know beforehand where the jewels will be located ● With this insight, jewel detection is as simple as comparing the sum of the red/blue pixels in these regions
- 13. Blurring ● A pretty common operation outside of computer vision for editing images in general ● Blurring is also common in image processing to smooth out images and remove noise
- 14. Thresholding ● Thresholding turns a grayscale image into a binary one ● Simple procedure: pixels below the threshold become black, pixels above the threshold become white ● RGB thresholding applies the operation to each channel separately
- 15. Alternative Color Space: HSV ● Alternate way to represent color ● Three channels/dimensions: ○ Hue ○ Saturation ○ Value ● Often used instead of RGB for color thresholding
- 16. Binary Morphology Dilate Erode Close (Dilate ⇒ Erode) Open (Erode ⇒ Dilate)
- 17. Contours ● Basically boundary curves ● OpenCV can find the contours that enclose regions in a binary image ● Allows for the pipeline to extract useful information from the image data
- 18. Example: Basic Cryptobox Vision Original Blur HSV Threshold ContoursFinal Binary Morphology
- 19. Motion Control ● Speed + Accuracy ● Lots of subsystems need motion control ○ Drive base ○ Jewel arm ○ Dump bed ○ Dump lift ○ Relic claw Acquire ActThink
- 20. Control Systems ● Get from state A to B ● Maintain a setpoint ● Various applications ○ Motion ○ Temperature ○ Chemical processes ○ Power supply
- 21. Basic Example Bang-Bang control Target: 72℉ If temp is under target, heat Otherwise, do nothing Overshoot Oscillation
- 22. PID Account for current error ● Slow down as it nears the target Account for past error ● Compensate for constant forces ● Gravity, Friction, Heat loss Account for future error ● Dampening ● Prevents overshoot
- 23. Motion Profiling ● Observe constraints ○ Max Velocity ○ Max Acceleration ● Time optimal ● Feedforward vs Feedback
- 24. Simple Algorithm ● Acceleration limited ● 3 segments ○ Acceleration ○ Cruise ○ Deceleration Calculate max Velocity Determine Accel/Decel Distance Calculate Cruise Distance Calculate duration for segments
- 25. Jerk Control ● Jerk - the derivative of acceleration ○ The rate at which acceleration is changing ● Acceleration Control necessitates infinite jerk ● 7 segments with constant jerk
- 26. Putting it all together ● Feed forward control - applies control in a pre-defined way ● Feedback control - applies control in response to the state of a system ● V and A terms - relate desired motor performance to voltage applied ● Lower PID gains - feed forward ensures that there is relatively low error
- 27. Follow Our Progress ● Website: goacmerobotics.com ● Facebook: fb.com/acmeroboticsinc ● Instagram: @acmerobotics Contact Us ● Email: goacmerobotics@gmail.com Learn More About FIRST ● Website: firstinspires.org Robot Code ● GitHub: https://github.com/acmerobotics/relic-recovery
- 28. Program Future
Editor's Notes
- #3: Founded 3 years ago by Ryan and Kellen because they wanted to do robotics and there was a lack of STEM activities in the area Grown to 11 team members and 5 mentors (coaches) Only brought the three senior members / team leads Community team (room at NU, but self-funded team)
- #4: International competitive robotics organization founded almost 30 years ago by serial entrepreneur Dean Kamen Focused on building skills for future business and technology leaders 4 levels of programs, starting for kids age 6 through high schoolers
- #5: Way more fun than the Federal Trade Commission We compete in the third level, First Tech Challenge, for middle schoolers and high schoolers Program is designed so teams run like small tech start-ups Team is responsible for creating and setting the budget and raising money for the year Team is split into multiple sub-teams: hardware, software, and business Documentation: engineering notebook that we present to judges at competitions, documents our team’s journey and design process Scheduling: team must figure out how to get everything designed and built among multiple sub-teams by a deadline Marketing: team branding, website/social media, fundraising Specific examples - like current budget tension between being over budget and wanting to rebuild the robot Community outreach: summer camp, school presentations
- #7: Talk about each sub-component of the hardware (and how we do CAD)
- #8: High-level overview of robot (how it talks to driver station and is controlled)
- #28: Up while Q & A