Welcome to project UGCSbot!

The goal of this project is to construct a robot which can maneuver around the UGCS lab at Caltech and deliver printouts to the appropriate user.

Team members

• David Stafford, Electrical Engineering and Computer Programming.
• Rohit Thomas, Navigation.
• Laura Rogers, Mechanical Engineering.
• Alan Somers, Accelerometers (no credit).

Original project proposal
Detailed project description
Software
project schedule
parts
Final report as of 3rd term 2002-2003
Contact: robot@ugcs.caltech.edu

Progress reports for Fall 2003

• Selected and purchased on board computer.
• Designed new drive system.
• Mounted and tested new motors.
• Re-resetup electronics
• fixed AI bugs

Week 2

• Setup on board computer and built its power supply
• Found a bug in the charging circuits
• Integrated all electronics and motors and computer nto the frame
• Fixed bugs in the control software
• One of the voltage regulators self-destructed; the temperature sensors are offline until it is repaired.
• redesign: mounted the computer with the electronics and put the battery below the electronics

Week 3

• Integrated axels, gars, and tachometers, and bearings for the drive system.
• Finished the main loop
• Learned much prolog
• Fabricated wheel mounts.
• Designed (at a high level) the new AI system.
• Finished the outline of the new vision code

• Negative progress. A screw got loged in the motor controller board's power regulator. This melted the insulation on the 5 and 12v lines, shorting them together. Smoke and flames ensued. Repairs are underway.
• Repairs complete. The Sensors board was unsalvagable, I will rebuild it when I actually need it.
• Finished system integration: the drive system, electronics and batteries are now integrated onto the robot. It can move reliably under its own weight
• Designed and started coding the vision system.
• Due to this weeks failures, I was unable to integrate the mainloop
• Started calibrarint the tachometers

• Integrated and callibrated the left tachometer. Something is wrong with the right tachmometer, preventing me from callibrating it. Mainloop testing cannot commence until both tachometers work.
• I've implemented enough of the vision system to realize that it has some serios problems. My approach was to apply the hough transform for a circle to the 1000 most significant points of the magnitude of the red-normalized gradient of the image. This fails because finding clusters of interesctions of 1000 hyperplanes is too difficult. Using a 3-point hough transform fails because of exponential blowup (1x10^9 points). I was considering using a two point Hough transform and assuming that each point was on the diameter of the circle. This method yields too much noise.

  I have developed and am currently implementing a new algorithm under the assumption that a particular edgepoint is only a member of a single circle. Although this asusmption does not always holds, it usually holds which is sufficient for these purposes.

  1. Let G be the magnitude of teh gradient of the red-normalized image.
  2. let P be a randomized list of all points in G with G(x,y)>5*stddev(G)
  3. Let P' be the inverse image transformation of all points in P about P(1). This transformation make all poitns cocircular with P(1) colinear with P(1)
  4. Apply the hough transform to find the most significant line in P' that includes P'(1). Let the set of detected points on this line be L'
  5. computer the inverse transform L, of L'.
  6. Add L X, the list of possible circles.
  7. remove L from P.
  8. If there are more points in P, go back to step 3. Otherwize continue
  9. sort X (a set of a set of points) by the size of the sets
  10. let C be X(1).
  11. let (x,y) be the average point of C, and R be the average of |C-(x,y)|. Scan subimage enclosed by the circle (x,y,R) if the average red-normalized pixel is high return(x,y,R).
  12. remove the first element from X and go back to step 10.
  I have implemented and am currently fine-tuning the algorithm. I do not yet have solid evidence that it works.
• Charictorized noise problems from the camera's compression algorithm. The compression algorithm inserts significant noise in the 2nd derivertives of the image, making steriovision by linear convolution impossible. I'm considering other techniques (corner classification) as well as more sophisticated cameras.
• Integrated new batteries.
• Performed a realistic terrain test. The robots reliability is good, but has serios traction problems. I'm buying new casters which should solve that problem.
• Found (but not root-caused) a bug in the DAC system in the IO board

• Rebuilt and installed, and callibrated the right tachometer.
• Tested the mainloop but my local path planning algorithm is faulty. I'll have to redesign it
• Finished the deadreconing software
• Worked more on the vision system, but its still not complete
• Added reinforcement bars to the robot.
• Setup the new power supply
• Installed the new casters. Mechanical performance has improved a lot.

• Finished and debugged the mainloop/local pathfinder. Although the principal behind this code is simple, gettingn it right is a lot harder than it sounds but it is working now.
• Integrated the robots case, neck, antenna and fan.
• Designed slip-detection software.
• Characterized EMI and power consumption.
• Added timing information to the low-level C code.
• Tested the servo motors on the neck.

Progress reports

• Cardboard mockup completed, but the robot is too low to the ground.
• The parallel port and ADC board was constructed a long time ago and still work.
• New microcontroller based motor control output sections built. (feedback sections not constructed yet). Only channel B tested.
• Debugger board for the SX28 microcontrollers built, and tested.
• Navigation literature survey completed.

• Motor control board feedback hardware built and tested (both channels). feedback software designed, but not written. Need to address some issues with the current sensors.
• Accelerometer selection and basic design completed.
• Dremmel tool broken while on loan.
• Laptop selection complete.
• Addressed the problem with the robot being too low to the ground. (didn't update mockup).

• Power supply and battery charger completed and tested.
• Aquired materials for robot construction.
• Dremmel tool partially repaired. Need to order new brushings.
• Designed the sensor board.
• Aquired LCD display panel and datasheet. Started work on an LCD test board.

• Sensor board built tested and debugged.
• Fixed a bug on the debugger board.
• Servo motors aquired, power train plan completed.
• Built the electronics bay and mounted the boards in it.
• Case construction started
• High level software design completed. (See the new software link)
• Got more hardware.
• Designed the task-manager. (still trying to prove correctness)
• Completed navigation code rough outline. (This will be online shortly).

• Designed and started construction of the UI/servo/reset board.
• Proved incorrectness of the task-manager.
• Measured the torque/speed curve for the motors.
• C/Lisp interface code written. Some debugging left.
• Fixed a bug in the motor microcontroller code.
• Preliminary drive system testing.
• Started some documentation.

• Frame construction complete.
• C/LISP interface debugged.
• Servo motor, laser, and emergency-stop hardware and software complete.
• Fixed a bug in the computer interface board.
• Drive couplers started.
• Feedback microcontroller code and dead-reconning code written.
• We decided to scrap the LCD display.
• Laptop purchased and setup. Parallel port compatability problem fixed.

• Quick and dirty versions of the mainloop, navigation, control and scheduler code completed. But not yet tested.
• The world tracker module has been integrated into the navigation system.
• More mechanical parts purchased.
• C/LISP code now avaliable here . Microcontroller code comming soon.
• Drive system complete and integrated onto the robot.
• The robot appears to be able to move it's own weight.
• Vision system designed.

• Decided to rebuild the drive system.
• Found bugs in the original framegrabber code.
• Wrote and tested image pre-processing and feature detector code.
• Wrote gradient descent and image transformation functions (untested).
• Decided to rewrite the entire AI subsystem.

• New AI subsystem designed. About 15% implementead
• Wrote more documentation.

• Wrote supervised learning code.
• Wrote more documentation.
• Finished camera neck.
• Started callibrating the camera (This takes a LONG time).
• Fixed bug in the servo motor circuit.
• New AI system about 40% completed.

• Fixed bugs in the camera code
• Presented the vision system to the CS148 class
• Integrated and talibrated the tachometers.
• Integrated motor temperature sensors.
• Fixed bug in the motor controller.
• Found bug in the sensors board.
• Nem code posted here .

• Build DC motor test board.
• Finish vision code.

• Navigation system desing and implementation.
• Finish ALL mechanics and electronics.
  • LCD display. (microcontroller programming)
  • E-stop system. (microcontroller programming)
  • Charging station.
  • On-board computer. (buy it)
  • Laser power supply. (built, but deosn't work for some reason)
  • New drive system.
  • Bump sensors.
  • Camera mounting.

• First 2 weeks: finish hardware and some software integration issues.
• Next 7 weeks: Finish all remaining computer software.
  • 2 weeks: write main loop and realtime feedback code.
  • 2 weeks: finish STRIPS algorithm.
  • 1 week: write reduction rules.
  • 1 week: state processor code.
  • 1 week: software integration.
• Last 2 weeks: Final testing, documentation, and demonstration.

Pictures

Electronics:

Top left: microcontroller debugger board.
Top right: Computer interface board.
Middle right: Motor controller board.
Bottom left: Optical sensor test unit.
Bottom right: Power supply board.

Finished electronics board

Robot base with casters and completed drive system.

Camera with neck.

Motors, computer, frame, and electronics all integrated.