My current project: a Large rover! I have found a cheap dis-used electric wheelchair on eBay, removed the "chair" and all of the electronics, created a power distribution system behind the battery pack based around an arduino Uno, two 8 channel relay modules, and two 15A step-down converters for 12v and 5v.
On top I have replaced the chair with a aluminum frame where most of the 'higher' level electronics reside, currently this comprises of an Arduino due, a raspberry pi, 1TB Hard drive, a pair of 25A motor controllers, a uBlox GPS, a network hub, WiFi bridge and a USB Hub.
The plan is to use the Arduino DUE as the brains for the motor drive, it's managing the navigation and drive of both motors. I plan to base the navigation around the 9DOF sensor pack I have added as a primary source of information, corrected by GPS & barometer for altitude. I have a set of SR04 ultrasonic sensors that will provide a 'distance' measurement for navigation as well.
Once the basics have been sorted (ie navigation & drive) I have plans to interface the rover into a Kinect sensor; I have begun work on this already, however I require a fairly powerful computer for the image processing if I want to do 3D point cloud, navigation (damn hard), and person recognition as security for commands (nearly done).
For now, I just thought that I would share...
Comments
Hi Jesse,
The UDOO Quad looks like a very good controller for a lot of applications.
For real time applications almost anything is better than Intel architecture and Microsoft operating systems.
The only thing they have going for them is that they are big and fast and relatively cheap.
Microsoft really gave up on RTOS capability when it left DOS for Windows and they have been begrudgingly and badly supporting any real time operations only as an unpleasant afterthought ever since.
And Intel architecture has been designed around supporting the Microsoft personal computer (non-RTOS) world ever since.
I might follow suit on the google forum...
Gary, I am considering using a laptop for processing the kinect stuff; the uDoo should be powerful enough to do it tho, ARM's are actually a lot more efficient at processing data than an Intel based processor, but I imagine that sooner or later I will want to use something more powerful.
with this in mind I'm actually designing the robot to take commands from a command queue on an SQL Server, this enables me to have a very common easy to use protocol to control the robot so I can plug my laptop into the onboard hub to deliver high level commands. all the low level stuff will always be looked after on the due; but nav commands will be able to be scripted from higher up the chain.
Thanks for the Google forum link Bot Thoughts.
We have had a previous Blog post on the Pixy, could be useful as a supplemental information acquisition device.
I ordered a Pixy.
Any of you doing anything on localization besides GPS/dead reckoning?
We're having a discussion on some options on diyrovers google group. RF backscatter is an interesting concept that started the topic. I'm leaning more towards basic machine vision solutions.
If you're looking for serious in-depth *rover* discussion with some very smart folks on algorithms, development, implementation, diyrovers is the place to be. We roll our own on diyrovers :) Mostly RoboMagellan and AVC competitors and others with various applications.
https://groups.google.com/forum/#!forum/diyrovers
For general robotics, the Parallax forums have a pretty impressive braintrust. Mostly I hang out on the Propeller forum but I think there's some general discussion that goes on. It seems like there are a lot of forums but it ends up diluting the robotics community quite a bit.
Have you guys seen the Pixy kickstarter? I jumped on it and should be getting a couple of units in November. There is only a few days left.
http://www.kickstarter.com/projects/254449872/pixy-cmucam5-a-fast-e...
Actually I haven't found too many forums, there is a lot of Arduino stuff.
And a few places like the Robot Store that are fairly serious suppliers for smaller bots, but not much organized for more serious bots yet.
With the APM and now the PX4 and PixHawk and coming PX CAN Controllers and our already existing APM:Rovers here, DIYDrones is probably well positioned to move into more serious actual Robots even though so far their emphasis has been on Rovers.
The reality is that the same stuff that makes the Rovers work well serves as an excellent starting point for Robots as well.
First you need a good mobile chassis able to navigate with absolute positioning and some additional simple obstacle avoidance stuff. And we are already pretty well there.
Then, the hard one, relative positioning and navigation based on the stuff in your environment.
At some levels this can be handled with things that would work with the PX4 derivative, but reality is the easier approach is with PC stuff with quite an existing hardware and software base out thereand the raw CPU speed helps for doing things like real time SLAM from 3D point clouds.
For our ground based bots, weight and size aren't so much of an issue
Jesse, A thought about the vertical control on the Kinect.
Basically the built in one is "totally" inadequate for dynamic adaptive vehicle use.
Mostly you want to keep it level at all times and that motor isn't designed to do that it is slow as molasses and designed just for very slow repositioning to where your head and hands are.
Modern brushless gimbals can maintain level perfectly without chattering (like servo style camera gimbals).
And you can still feed in an external PWM signal to allow it to primarily point at another angle while still having the controller provide perfect stabilization.
In robot image processing the most important thing is to NOT have your image bouncing around including for 3D point clouds.
The stabilizing platform is critically important and the built in Kinect tilt is completely worthless for it.
Basically throw it away.
When the Kinect 1 comes out it will probably up resolution and discrimination performance by 3 or 4 times over the current Kinect (although, unfortunately, not range).
TOF is way better than structured light for this.
Hi Jesse,
I should have clarified, the brushless wheels are 48 volt, and I am using a 48 volt DC to 5 and 12 volt DC switching converter for most of the electronics.
The lap tops voltage regulator does come off of 1/2 of the 48 volt pack, but the power imbalance it will cause is so small as to be negligible.
Heat really isn't a problem, because the wheels are designed to dissapate 500 watts under normal operation and will spend most of their operational life averaging less than 200 watts (LOTs of aluminum in the hubs).
I have 2 Kelley 48 volt brushless motor controllers that are compatible with my wheels and which use simple analog 5 volt inputs for throttle and a single digital bit each to enable regenerative braking and reverse. (reverse is the trick, otherwise I could have used the controllers that came with the wheels.)
Everything else is 5 volt or 12 volt except for the Laptop charging regulator and that is covered by the 48 volt converter.
Hi John, Right you are, swinging forward and back is a characteristic of the hanging pendulum.
And it will swing forward and back as you accelerate / decellerate (apply wheel torque in any direction).
My vision system is mounted on a brushless gimbal to compensate and acceleration and deceleration will be moderated to keep this effect within bounds.
But it is in static balance unlike a conventional pendulum 2 wheel robot or Segway which actually align the angle of the frame in accordance with their actions.
These days with modern gyros and accels it is simply a choice as to which works best for you, I simply chose the less traveled road.
Hi John, I haven't actually found a decent website besides this one; I posted my robot build on here because I've historically done a stuff on my hexacopter, and I recently decided to build a rover based around some concepts that I had while playing with the arducopter stuff, followed shortly after by a few ideas with a Kinect sensor and the Udoo boards. my ultimate goal is to write a core platform on the rover (they don't fall out of the sky and smash when you make mistakes!!) on the arduino due board, based around the concept that the IMU controls everything as a primary source, and the GPS, and other sensors are used as a point of reference.
Once I have the core sorted out I plan to implement navigation via the kinect sensor, and afterwards fly this configuration with the udoo board... that is the plan at least. this was the best forum to post my updates in for that reason. Plus I know that the DIY Drones people would have good feedback and would have been interested!!
Gary might have an alternate point of view that could enlighten both of us when it comes to forums?!
Both of your robots are really nice. I have a question for Gary, how much swinging forward and back does your frame experience and does that affect your sensors? Did you consider an inverted pendulum?
I am also into robotics. Have you guys found any websites similar to this one, only geared for robotics?
-
1
-
2
-
3
of 3 Next