August 25, 2005

Sensor Mounting Techniques

Sensor Mounting Techniques

Introduction

While upgrading one of my Mini-Sumo robots, I found that sensor placement and orientation can have a big impact on your final results. This article will discuss the problems I ran into, trouble-shooting, and solutions.

Time to Upgrade

Mini-Spat has competed in 6 PAReX events, as well as several demos at swap meets, high schools, and colleges. In other words, there's a lot of mileage on the little guy. During the Nov. 2003 event it died twice in competition, and in the Nov. 2004 event, there were a few times when it lost it's mind and wandered the ring aimlessly. The original circuit board was a low-profile wire-wrap board. With the extremely short wraps I was concerned about it's reliability and decided to upgrade to a PCB.

Upgrade Woes

The PCB was finished and mounted to the robot. There were no software changes made, except swapping the left and right servos, due to PCB routing issues. Fired up the robot and watched it act just like it did with the old board, in other words, a robot possessed. Since the PCB seemed to be correct, and nothing else had changed, I started looking at what I had done in software, trying to figure out what went wrong.

Basic Trouble-Shooting

The basic problem was the robot was seeing things that weren't there. When Mini-Spat drops, it's suppose to move forward, then start spinning in a circle looking for the opponent. If there's nothing in range, it should just keep spinning. Most of the time it wouldn't even make one full turn before wandering off in search of a ghost. Other times it would work fine for several runs.

By inserting some breakpoints in the code, I would stop the robot when it thought it saw something, and flash the LED to tell me which sensor was acting up. Turns out the right range sensor, a Sharp GPD012, was at fault. Further investigation showed that two of the three connector pins had broken solder joints! Ahh problem solved!! A quick fix with a soldering iron and it was back on the ring, failing exactly the same way. Insert your favorite expletive here!

Digging Deeper

Well at least I found the intermittent problem. Most Mini-Sumo's don't have a display on them so it can be hard trying to figure out what's going on sometimes. I have a diag routine which can blink an LED to show me an 8-bit sensor value, but you can't use it while the robot is running. I guess the ideal interface would be a wireless link to a PC, then you could receive all kinds of data to see what's going on. Well those links aren't cheap so I fell back on the RS-232 link that I had brought over from the original board. A long 2-wire cable connects the robot to the laptop, but it's still not practical, even on the 30" ring. Since the failure was a range sensor, I did all kinds of tests without the motors running, and the sensor performed perfectly.

Ok, since this is an analog sensor, maybe the motors are generating some electrical noise and hosing things up. Putting the robot on blocks and running the motors during the tests, made no difference, the sensors worked as expected. The only thing left was to monitor the sensors while the robot was running normally. To do this without the cable attached, I would save the sensor readings to RAM, and then connect the cable and dump the data to the laptop.

I had about 250 ram locations to use and as fast as the sensors are updated, it would fill in about 38ms. So I once again set the robot to stop when it saw something and send me the last 250 readings. What I saw was quite interesting, the left sensor read a 0 or 1 the entire log. The right sensor varied from 0x20 to 0x28, before hitting my 0x2C threshold and stopping. The 0x2C is about the length of the ring, so I would "see" things in the ring, but didn't care about what laid beyond. If I did the same test with the motors off I got readings just like the left sensor, nothing greater then 2. If I ran the motors on blocks, again the readings were good. I changed out the sensor and again had no changes. So the actual motion of the robot had something to do with it.

Sensor Orientation

Somewhere along the line a thought hit me, the only difference between the two range sensors was they were mounted 180° from each other. This was done so both connectors faced inward and kept the wiring neat. In case you're not familiar with this type of sensor, it has an IR emitter on one end and an IR receiving matrix on the other end. The right sensor has the emitter on the down side, shooting upward, while the receiver is on the upper side looking slightly downward.

Here is the orientation of the sensor on the left side of the robot.

And here is the sensor orientation on the right side, the one acting up.

Could the sensor be seeing reflections off the actual arena surface? There was just enough extra wire to allow me to rotate the right sensor such that the receiver was now on the bottom side, looking slightly up, just like the left side. A quick run and a log dump showed perfect readings, no noise or junk, or whatever was going on.

Apparently whatever the right sensor was seeing would only occur while the robot was moving, probably a reflection of overhead lights or sunlight through the windows and off the arena surface, even though the surface is flat black.

Conclusion

Rather then leave both sensors mounted vertically, I think I will make a new bracket and mount the sensors horizontally like I did with FlatSpat. In this manner the sensor won't see anything from above or below and hopefully be more reliable.

How did it work for 5 or so events without failing? A look back over my code changes shows that at some point I had changed the range threshold from 0x30 to the current 0x2C. I suspect the reflections rarely gave a reading greater then 0x30, so I never saw the problem. There was one time we had it running outdoors at a swap meet, and it did act up a lot. May have been the same problem.

So at the next event, look for Mini-Spat's new eyes, it will be looking for FlatSpat, Zorro, and Mr. Happy!!

Posted by Kelly Small at August 25, 2005 10:16 AM