This is version 1 2 3 4 5 6 7 . It is not the current version, and thus it cannot be edited.

Back to current versionRestore this version

The following is some advice provided me by David Anderson of the Dallas Personal Robotics Group (DPRG), which is what he called a "build sequence". It's basically a plan of action that will help anyone focus on getting their robot going, step by step.

Here's David's advice (edited slightly for context):

---

For beginning robot builders or even for experienced builders, I've often suggested a build sequence like as unto this:

1. Build or buy a two-wheel differential drive[1] platform.  Get a development system working and write enough code that you can set the motors' PWM[2] at various constant velocities.  Don't worry about encoders or PID or behaviors.
2. Run the robot.  Does it run in a straight line, or veer left or right?  Can you make it run in a straight line by adjusting the two motors to different speeds?  What is the slowest it can run? On a hard surface? On carpet? What is the slowest it can run and maintain a straight line?  Can it climb over things?  Like a doorway threshold or a book?  What about with one wheel on carpet and one on a hard surface?   It is good to know what problems encoders and PID are intended to solve.
3. Build or buy a bumper and get a bumper behavior working.  This is the "sensor of last resort" that will save and protect your robot when all other sensors fail or misread. The classic bump-and-run algorithm is a good place to start. If bump is on the right, stop, back up, turn left a bit, go forward a bit, release control. Same if bump is on left. If bump is in center, it's not so obvious what to do, so it invites creativity.
4. Run the robot. A lot. The robot should be able to survive on its own in any environment.  Watch it and see how it does. Does it get stuck and require human intervention? How could it recover from the situation on its own?  What information does it need to know it's stuck?  What should its response be? How can you solve the stuck-in-a-corner condition of the bump and run algorithm?
5. Once you are confident the robot can survive on its own, start thinking about other sensors. Photo sensors, (basically just a cadmium sulfide cell as half of a voltage divider connected to an A/D input[3]), in a pair pointing forward, angled slightly left and right, is an easy sensor/behavior to get working, and the results are quite satisfying. Turn toward the brighter side, perhaps with a deadzone in the center. The robot can seek toward bright lights, which change with its position, so it can apparently navigate a space quite intelligently.  Or seek away from bright lights (turn away from brighter side). This plus the bumper should give you a robot you can play with, in any space.
6. Run the robot.  A lot.
7. Now start looking at avoidance sensors. There are lots of ways to do this and more than one technique will probably be required for robust behaviors. I like IR proximity sensors[4] with a wide FOV [Field of View], again in left/right pairs.  The behavior is pretty simple: just turn away from a detection.   But lidar, sonar, IR rangers, ping sensors, all are good things to experiment with. And as mentioned, robust behavior will probably require some combination of sensors.
8. At this point you should have a robot that can be turned loose in any space, mostly not hit things, recover when it does, and navigate intelligently around the space based on the the room lighting.

Now time to start thinking about higher level behaviors.  So that means going back and implementing encoders, PID, and odometry. Those things open a whole new world, with its accompanying complexities, so it's good to actually have a functioning robot before getting to that stage. And if one never does get to that stage, you still have a functioning robot you can play with and show off to your friends.

Hope this is useful.

cheers!

David

Footnotes#

---