NXTSegway Revision 0.4 Final: New Design (Again) and PID Controller!

This is the fourth public release of my NXTSegway, and I am quite pleased with the results. I re-re-designed the robot: Dean Hystad informed me on the NXTStep forums of the benefit that a higher center of gravity can give to a balancing robot like NXTSegway. He gave an excellent example: it is much easier to balance a golf club on your finger than a pen. Along with the bigger wheels, as the title says, I also implemented a PID motion control system that solved all my balance problems. The program uses three error components (kp, kd, and ki), multiplies them by three constants, adds them together to form the PID value, scales the PID value to fit in the motors' power range (-100 to 100), and sets the motor power equal to the scaled PID value. The three error components work like this: kp (proportional) changes proportionally to the error, kd (differential) changes proportionally to the rate of change of error, and ki (integral) changes proportionally to the sum of all previous error values; the constant multipliers weight these components based on need. For a complete, technical PID tutorial, check out this site. As you can see from the video below, this controller is highly effective in creating a stable inverted pendulum. Note: if you want to use my source code and have a working robot, you must use a very similar design and adjust the kp, kd, ki, adj, and scale values to fit your robot and its environment. My values (especially the kp value) are extremely high due to the fact that the batteries were low...once I replaced the batteries, there was WAY too much overshoot.

Annotated source code can be found here (4KB)

NXTSegway Extreme Makeover

Well, as cool as the old NXTSegway looked, with the big wheels and all, I decided that it is time to redesign it for several reasons: 1) The big wheels increased the height as well as the center of gravity, making it harder to pull the robot back to equilibrium, 2) The big wheels actually caused the robot to over-correct and lose precision due to the much larger circumference, and 3) the larger wheels prevented me from putting a bar across the front to hold the motors on laterally, so I had to settle for strapping a green rubber band around the middle, which kept getting in the way. I am going to continue the NXTSegway project with this model, which will hopefully yield better results. If you want a model of it, you can download the .lxf file (Lego Digital Designer file) here. I'll post pictures or video with the release of the next revision.

NXTSegway Revision 0.2

OK...this project has experienced a major breakthrough today, as I have implemented a form of dynamic recalibration of robot's "stable point." As you can see from the videos, the robot moves back and forth, and I was able to use that movement to my advantage. The new program measures the raw light data at the extremes (at the point when the robot is momentarily stopped and ready to move back the other way), finds the average of those extremes, and sets the average as the new target light value. This way, if the robot begins to lean a tiny bit too much in one direction, the target light value is shifted in that direction so as to maintain balance (ideally). However, with this method, there is still no way to overcome any substantial error, only minor error, so if it leans too far it will careen into the ground (as evident in the video below). If you want further explanation, take a look at the commented source code and see if you can give me any suggestions :-)

Source code can be found here (4KB)

NXTSegway Revision 0.1

This is the first revised edition of my project NXTSegway, which was first released yesterday. As you can see from the video, the robot is a tad bit more stable and, with some extra work in the next couple of weeks, might just rival some of the other segway-bots out there :-) To achieve quicker response time, I replaced the scaled-to-100 light sensor data with the raw, more precise 0-1023 data. I tried to implement some sort of bias into the correction process to overcome the imbalance in the structure itself, but after a couple hours of fruitless work I found that if I calibrate the robot's "vertical" light value while tilting it slightly, the program will compensate for the imbalance without any extra code. Granted, this will need to be changed, but it works for now. I plan on adding some sort of dynamic adjustment of the midValue later this week; it is harder than I had previously thought. Anyway, here is the new video and code:

Source file can be downloaded here (5KB)

My First Project: NXTSegway Initial Release

Now, before I write anything else, I know this is nothing new, exciting, or revolutionary. However, this idea is what prompted me to get back into Mindstorms - I saw a video of one on Youtube and thought it was cool. This is my first attempt at a self-balancing, two-wheeled robot in ROBOTC (I couldn't do it in NXT-G) using a light sensor. To run it, one must first calibrate it, storing the values of the light sensor when vertical and when tilted to both sides. Then, the robot reads the value of the light sensor continuously, calculating the direction of error and compensating by using motors to rotate the wheels in the direction opposite the error. It is extremely jerky and unstable due to its lack of complexity (I wrote the program today just to see if this simple idea would be enough to stabilize the robot). My next revision will probably include dynamic adjustment of the midValue, which determines when the robot is vertical, because in its present state the robot moves around a lot and the midValue is not adjusted accordingly (the amount of light directly underneath the robot changes as it moves around). My ultimate goal is to purchase a Gyro sensor from hitechnic.com and implement some sort of motion control system (probably PID) that would produce (ideally) a very stable robot. Ryo Watanabe did something like this and his project, NXTWay-G, is documented completely at his website. Anyway, here's a video of my sorry robot and the code that runs it:

Source file can be downloaded here (4KB)