I am planning to build a quad and control it autonomously by using monocular camera(nexus 5) and a kinect sensor. I know it is a very ambitious project but i have some experince with computer vision so the main challenge for me will be the building the quad.

I am a beginnier with flying robots but i have built a few land based wheeled robots. My main aim here is not actually building a quad but acquiring a solid platform that i can use for my research. I open to all options but what i am considering right now is building using an existing frame and adding motors, esc, controller etc. I am also open to RTF or ARF kits but they should be open enough that i can easily change the code on the controller to suit my needs and control it vision and the kinect. 

My plan is to build the project in three phases starting out with a very basic build and slowly adding features.

Phase I:
1)Build/acquire a radio controlled quad that can fly stably
2)Add a gps for autonous flight through pre defined waypoints

Phase II:
1)Add the nexus 5 to the equation as an onboard camera with on board processing using opencv for android and use it for simple obstacle detection in conjunction with the IMU and the GPS.
2)Use the camera for indoor navigation with help of ultrasonic/IR sensors in GPS denied environments.
3)Use the monocular camera for fully autonomous SLAM and navigation without IR or sonar

Phase III:
1) Add the kinect sensor for even better indoor navigation.
2)Use the nexus for capturing ground images and processing and kinect for forward object detection.
3)Use the combination of kinect+nexus for outdoor and  indoor navigation using all the sensors and GPS for completely autonomous flight.

I am soon planning to begin building the quad and have a few questions:
1)Which flight controller would be the best here? I was thinking of APM 2.5
2)The major parts i am planning buy for the  quad:
    ESC: Turnigy plush 30A
    Motor: NTM Prop drive 28-30 900kv 215W @ 11.1 v
    Frame: Hobbyking q450 glass fiber with pcb
    Battery: Turnigy 3S 2200mah 25c
  Hope these will be good enough for my purposes.
3)For the radio transmitter and reciver is it better to buy locally or order from hobbyking?
4)Same for the other parts which parts would be better if imported from hobbyking or rctimer etc. and which parts would be better if bought from india online?
5)I already own the 2200mah lipo battery would i need a beefier battery for about 10 minutes of flight time, is the discharge rate of that one enough or i need a 4500mah nanotech? Also can lipos above 4500mah be imported as i read about some issues here.
6)My budget for the quad is around Rs. 25,000/- is this enough for everything including the accesories,sensors and radios etc?(excluding the kinect and the nexus obviously)

Thanks in advance for reading through this long-ish post. Awaiting your replies.

PS:I will be using fedex/dhl for courier if i import.   

1) APM 2.5 is good, its open source so you have all the code out there. Altho there is Pixhawk, which is fairly new has lot more processing power, redundancy. It uses the same firmware as APM. Im not sure if kinect+nexus can talk with Pixhawk/APM, so I hope you would know the finer details. I think you can also use a beaglebone with an IMU but it hasn't been done as yet, people are working on it.

2)The setup looks good,how much weight do you intend to carry? An approximate estimate will help to decide the setup.
Think a bigger frame will be good for your purpose as you would be mounting lot of items,which will require more real estate.
Also look for higher discharge batteries as motors in all will be pulling <80amps.
Instead of an X or + shaped frame you should look into "Deadcat" or H quad frame as they have more space.
http://www.hobbyking.com/hobbyking/store/__25884__Dead_Cat_Conversion_Kit_for_SK450_Quadcopter_Frame.html
http://www.hobbyking.com/hobbyking/store/__35908__Hobbyking_H4_Copter_Multi_Rotor_Quadcopter_Frame_470mm.html

3) You can buy locally if there is availability, www.quadkopters.com had some radios last time I checked.They are multirotors specialist so you will find most of the items there.Importing has its quirks like customs trouble or shipping time etc, sometimes they may ask for documents/approvals/licenses. Fedex/DHL will handle things better but they will charge duty for sure. Try to get radio locally if possible.You can check this thread for more info on LHS's
http://www.rcindia.org/rc-outlets-and-resellers/mega-list-of-all-rc-outlets-in-india/

4) Most of the things available in India are the same as HK or any other manufacturer. Very few LHS have separate manufacturers. You should ideally compare prices and check availability and order accordingly. Some products would not be available here so you will have to import them anyways.

5)Since you already have a battery, why dont you build up your quad and try it, if its not upto the mark then you can order bigger capacity batteries. Yes many shippers dont allow Lipo because of there dangerous nature but you can import them as there are some shippers who allow it with some more shipping fees. It shouldn't be much of a trouble.
C rating will depend upon AUW of your quad and the prop you use. Your current battery can discharge 55amps max. And the motor can pull from 10-20 amps each depending upon prop and voltage. A 3500-5000mah 20c 3/4s would be apt for the purpose, you can select the prop as per your AUW.

6) You will need to increase your budget a lil more I guess. A flysky 6ch programmable radio and APM 2.5(clone) will set you back upto 10k, then add telemetry/gps/ultrasonic/sonar/fpv/frame/electronics and then some. You will easily cross 25k. You can save some money on frame if you can built it yourself, its not rocket science can be built with some research. But in the end you will probably spend around 40k or more.

Im no expert just wanted to put some points for you to ponder upon and would love to see it at work, specifically the GPS free fully autonomous indoor navigation  .

Thanks for all the advice sidewinder. I considered the points you mentioned and searched a bit around the internet.

Most of the major research groups are using Ascending's Pelican UAV but it costs a bomb(6000 euro!!!). I also had a look at the specs it is 450mm in size and can have a custom built micro sized pc on board with intel i7 and 4gb ram on board. I can try to build something similar without the expensive bits obviously.

Another set of researchers are using the Parrot AR drone succesfuly for a lot of research. But the computations are perfromed off board on a laptop. This would be an option since it is cheap has two cameras already installed but i dont think i would be able mount a kinect on it. Parrot also provides an sdk for communication with the sensors and a lot of third party code is also available.

Building my own quad obviously would have several advantages but as you mentined communication with the flight controller would be an issue. Also i would have to use off board processing so i will need some sort of wlan connecting both of these. I can issue commands to the controller as they are PPM(if i am right) commands. The challenge would be getting the sensor data from the controller without much delay to realistically compute and issue commands in real time. I could try building my own controller using a better processor but that would be a project in itself.

I will have to look into APM a bit more to understand if what i am looking for is possible. Thanks for the help.

Well finally after a bit of searching around i have thought of going forward with the following:

1)For the flight controller APM is the perfect choice it seems now. It is completely open source and i went through the forums for devs and there is good discription on communicating with the Ardupilot using the standard MAVlink protocol using the UART ports. There is even an app called andropilot already developed that uses an android phone as the ground station. so the communication between the nexus and the apm might work out just fine.
2)The various other parts that i think would be appropriate are:
 1)NTM prop drive 800kv 300w BLDC motor
 2)Turnigy plush 30A ESC
 4)Q450 frame with integrated pcb
 5)ublox 6m gps
 6)turnigy nanotech 4500mah lipo
 7)10x4.5 props
 8)433mhz telemetry kit
 9)flysky 6ch tx/rx
3)I am planning to order the Arduflyer 2.5.2 with the gps from rctimer, is it a good idea to order from there?
4)Rest of the stuff from hobbyking, but strangely i was looking at the shipping options and it was showing fedex priority of $44 but on the payment page only EMS and airmail were shown.
6)Radio and telemetry kit online from india

My final concern and question before going ahead and ordering the above stuff is that will the above configuration be a good one for a first quad build? Or should i get both the kk 2.1 and Ardupilot, learn first with the simpler kk and then put on the ardupilot? Though i think there are enough tutorial videos and posts for both them. How difficult is it to actually build a quad the first time using the components i am planning to use?

I would really love to hear the thoughts of the more experienced members out here even if they are trivial since i am new to this. Also please feel free to share any thoughts on the above setup.
Thanks

I just completed my first build. I would recommend by both KK 2.1 and APM. I have already crashed couple of times so build is one thing and flying and tuning is another aspect. Order some spare components also if you are getting it from offshore.

Well i have recieved half of my components i oredered, one parcel from quadkopters, and one from rctimer. I ordered the parcels on 14th and already recieved it today!! Must say nice service from both, more so from rctimer as they ship from china. I had selected the express shipping option and it was shipped via DHL. Now waiting for the final parcel from hobbyking. I had selected fedex priority but i havent even recieved the tracking number yet. :/

A question:
What guage and type wires should i use for wiring the battery, esc etc. ?

Typically you wont need to extend the wires as they are long enough in most of the cases.
To extend wires from esc to motors or power distribution board to esc 16/15 awg should be good enough. From battery to power distribution board 13/12 awg. It all depends on current draw so you can read more @

http://www.rcgroups.com/forums/showthread.php?t=320240

Thanks a lot sidewinder for that link.

Recieved the final package from hobbyking. That completes all my material required. Will start building the quad soon.

The final list of components for my build is:
1)NTM Prop drive 800kv 300w motor(short shaft)
2)Hobbyking Q450 frame kit with pcb
3)Turnigy plush 30A esc
4)Arduflyer 2.5.2
5)KK 2.1
6)Ublox CN06 plus GPS+magnetometer
7)RCtimer AVdome anti vibration dome
8)Turnigy 5v UBEC
9)10X4.5 props
10)Turnigy Nanotech 4500mah 25-50c
11) FS T6 Tx/Rx
12)3DR 433mhz telemetry kit
Will post some pictures soon

great set up,will be watching your post,as i am also building a quad

A little setback on my project

I recieved the turnigy 9x today, it was the final component to arrive. Excitedly i unwrapped it and connected my lipo to turn it on and lo and behold, there was smoke. Turns out i had reversed the polarities. I opened up the case to find that a capacitor was fried. I'll try to fix it and post further updates.

It could be more than a capacitor. I had a similar situation once. My tx overheated and died due to use of dry cells.
If you post some pics, I could help. Also, I have a ton of electronics spares. Just ask if you need any.

Sure, on visual inspection only the 10uf seems fried. I am attaching the pic of the main circuit. Please feel free to have a look. And the components all seem to be SMD and i dont have any experience soldering surface mount.

If it isn't just the capacitor, reviving your transmitter will be difficult. Try changing it and see if it works.

If you don't mind me asking, what's your major?

It is computer science.

I looked around a bit on internet and there are more than a few threads where people have had similar problems with the 9x. Most of them were able to fix it by replacing the capacitor and/or the VR on the board and the VR on the transmitter.

Can i use a regular through hole electrolytic capacitor of the same rating? Do you by any chance have the replacement components?

Through hole should work. I guess replacing the VR and the cap should do it. In most such cases the damage pulse stops at the VR.

I do have the components. You can also get them at your nearest electronics shop, they won't cost much. If you don't, I'll be happy to help.

Just curious, what SLAM algorithms have you worked with? Have you implemented visual SLAM before? I'll be trying the EKF later this summer.

Sure i'll look for them tomorrow at the electronic market.

I have not worked with visual SLAM before. I did learn about kalman filtering on udacity, Here is the link for the course: https://www.udacity.com/course/cs373

PS: I found that nex robotics has the exact smd components, i think i can order from there http://www.nex-robotics.com/

Update on the remaining construction:

The rest of the quad assembled, soldered in the esc's and mounted the motors, mounted the kk 2.1 added the ubec. Everything powering up fine. The sensor tests on the kk show everything is fine. Now the only thing left is adding the props and flying. Now have to somehow get the radio to work so that i can program the esc's and actually fly.

Update on 9x repair:
Searched locally for components, did not get any smd components neither could i find a regular through hole 3.3v voltage regulator. Replaced the capacitor and the 5v regulator.

Completed my build today. I couldnt revive my Turnigy 9x due to lack of parts so i purchased a flysky t6 instead.

Right now i have the kk 2.1 mounted on my quad and i tried to test fly it today on my terrace it lifted off fine but it was a bit unstable and drifted to the front while take off and I had never flown anything before and i soon crashed it into a railing.   it wasnt a bad crash everything is fine only a prop broke off, but i have lots of replacements. I will take it to an open field tomororrow to and record a video. I have to tune the kk for a stabilized flight. I will post more photos soon. I will also post a detailed build log later.

I cannot express how great the feeling is when your creation first lifts off the ground. I am really glad i am off to a great start, now soon i will start with the later stages of my build.

Good build! Waiting for the maiden video.

I think you should have it tethered with some kind of strings while tuning the gains.

First flight of the quad ended in a crash. The quad was not very stable. It veers in a particular direction right from lift off. It also suddenly flipped in mid air during the flight and crashed fortunately i wasnt flying very high. Flying quads is not that easy.

I investigated why the quad crashed, it turns out one of the motor had a loose shaft and during previous landing it came a bit loose and during the flight it caused the motor to slow down flipping the quad. I need to get a 2mm hex wrench to tighten the grub screw that holds the shaft in place. Hopefully this will also cure the drift in the particular direction.

ayu we're building something very similar, a quad on kinect.. maybe we can jam together  and exchange our views regarding the same... feel free to ping me at 9555569902 
good day 

If your quad still drifts after you repair your motor you can try Auto level tuning.

Thanks a lot sidewinder that video really helped me a lot, the quad was drifting less after playing around with the auto level settings.

I have yet again hit another roadblock one of the motor again had a problem. This time it was a shot bearing, it again happened mid flight but this time i was cautious and miraculously caught the quad mid air as it was falling preventing any damage. I dismounted the motor and opened up the casing to find that the bearing was completely shot with the balls strewn across the motor. I cleaned out all the debris removed the shaft and the other bearing and as I tried to remove the damaged bearing to replace it, it came completely apart leaving only the outer race stuck inside. I tried everything i could to remove it without any success. I tried heating the aluminium and cooling the bearing but yet it wouldnt budge. I have to somehow pull it out but there is no give on the inner side. I tried cutting it using my dremel tool as suggested in a youtube video but turns out the bit i have wont even make a scratch on the hardened steel, turns out i need a tungsten carbide cutting bit for that. Also i couldnt order extra motors with the other stuff as i had exceeded the weight limit. Any other ideas that might work? Anyone know where i can get a 1/8th inch tungsten carbide cutting bit about 3mm shaft so that it fits my dremel?

Couldn't you just get another of those BLDCs from HK? Looks like it didn't take long to ship last time.

It didnt take long because i chose fedex that time, and it costs a bomb to get it shipped via fedex $39 to be precise. I had ordered a lot of stuff so that cost could be justified as regular shipping was around $25 but now for paying that much in shipping for $15 worth of motor would be unwise since regular shipping would just be $10. Plus you always have to pay duty when you get it shipped using fedex.

Hmmm... It's just that the BLDC might get even more damaged while you are trying to fix it.

Have you tried searching for it at the local stores?

Having strong techniqual strong base we develop high performance, reliable, and easy to use unmanned aerial vehicle (UAV), for commercial and recreational.
for more info visit my page: https://www.facebook.com/funaster

I managed to get the bearing out, but had to damage the motor a little, i made a small cut in the aluminium behind the bearing which allowed me to get some leverage to push the casing out. Installed new bearings, and checked the motor for any performance issues, so far nothing has come up.

The quad is finally complete now and is surprisingly stable. I am now going to add the apm with the gps instead of the kk.

In other news i have been offered a research internship at Robotics Reasearch Center at IIIT-Hyderabad for this summer. So i will be continuing my research there. Hopefully i will be able to make substantial progress over the summer particularly in the vision based navigation.

Please post if any further progress has been made. I am building a object following quad using APM 2.6 as a research project @ IIT Kanpur. So, the core part of my project is also the same as yours. The only difference is that I am planning to do the video processing off board. I need to control the quad using a C/C++ code. Any advice/comments would be appreciated

Updates:
The initial project i was planning has been pushed back a bit. Here at the IIIT-H i am working on a rudimentary ground station using MAVLink, where i will try to control the quad by using the laptop. This will then be integrated to fly the quad using hand gestures via a kinect.

So for i have written a basic console application that can read and send packets to the ardupilot mega via both the usb and the 433 mhz telemetry link over serial. The app is in c# right now and uses a lot of code from the mission planner. I plan to move to C++ as that is more familiar ground for me, but understanding mavlink is going to take a while.

@chawlap A lot of what you want to do would depend on what setup you have. If your quad is based on ardupilot mega or pixhawk i would suggest you have a look at the mavlink protocol for the communication. The mission planner is written in C# and is open source and so is mavlink, have a look at their code and try to figure out how to communicate with the quad. I can tell you what to do specifically for ardupilot as i am working on that.

@ayu135 Thank You for your reply!  

I did went through the working of Mavlink and ArduPilot Mega ( I'd also be using ardupilot   ) and what I inferred was that follow me mode would be the most suitable for object following and I can simply give the target coordinates of the point where I want my copter to go. But along with this, I would also need to control the yaw of my copter when I want it to change its orientation while standing at the same point. So, do you know of any methods by which I can directly control the yaw of my copter. Specifically, what commands would I need to send using Mavlink?

Any other comments/suggestions also appreciated.

Well you could look at the command rc_channel_override, what it does is overrides the rc input. It takes a value between 1000-2000 for each channel. A value of 0 for a particular channel means you dont wish to override it. You can have a look at the mavlink documentation for a complete list of all the commands available and how they are structured. You would need to write a program that will send the packets via serial for you.

You can have a look at the mavlink repository on github ( https://github.com/mavlink/mavlink ), clone it or download via zip and you will find a mavgenenerate.py in it. What this mavgenenerate does is it takes the xml definition file found in the message_definitions folder and then generates the appropriate headers in C,Java,Pyhton,C# etc.(you can choose the language) these headers contain all the enums and parse funtions for all the messages, it also has methods for encoding, decoding and sending messages using serial.

Follow me mode needs a special usb gps module and gps doent work indoors, and you'll have to mount a laptop an that object with this dongle attached. When you said object following i though it was through vision. Is your project on following object outdoors with gps reception?

Nopes! You had got me right the first time, I am working on vision based object following only.

What I was thinking was this : We can get the GPS coordinates and the direction in which our quad is facing via Mavlink. Right? We'd also be getting the live video from the quad through some wireless system. Now, if our target object is going to the left part of the frame of video (i.e. we want our quad to turn left while going forward) we can send 'fake' GPS coordinates (fake in the sense that we wave not got these from any GPS device)  which are front, left with respect to the quad. So, now the quad will try to go to this 'fake' GPS coordinate and in the process, it will turn left while going forward.

I am sorry if I am getting anything trivial wrong, I am a complete newbie and never got a chance to deal with this stuff. But does this sound possible to you? I would be happy to work on any other option also if it seems better to you.

@chawlap:
All you need to do is use a good object-tracking algorithm and send the appropriate control signals (in this case 'yaw signals'). You don't have to mess with the GPS system.

Another round of updates:
The work with quad is progressing, really enjoying the work here. Getting to learn a lot about quads in depth.

1)The ground station is now a gui based proper windows application, we can now control the roll,pitch and yaw of the quad using wasd keys,change the mode of the ardupilot, arm/disarm, it can read all the imu data from the ardupilot and display it, it can send set_roll_pitch_yaw_throttle command to the quad for more precise autonomous control of the quad instead of the rc override commands.

2)On the ardupilot side after fiddling around with the arducopter code a bit we found that the atmega processor is already pretty taxed by the existing code and adding any more functionality was not possible. We are now planning to rewrite our own controller from scratch using the ardupilot libraries as much as we can. Possibly we may be looking to move over to px4fmu+io or a pixhawk but that depends on whether we can accomplish it using the ardupilot after coding from scratch

3)After trying to get the position estimates using the gyro and the accel we realised that the errors are too large to be realiable after about 300ms of integrating. We are now trying ot get the px4flow sensor to work with the apm. The px4flow is indeed a nifty little sensor that might solve our problems to a large extent.

4)The gesture control using kinect was implemented though not optimally, the quad was too unstable to be controlled only via gestures. The gesture recognition on the other hand was impeccable. Once we get the quad to stably hover indoors using the px4flow the quad would be easier to control.

5)After we get the quad to hover stably in one place indoors witout drifting our next task is to get to follow simple coordinates sent via mavlink and navigate to them autonomously and then have the quad make basic figures like a square autonomously by sending the coordinates of the points.

Gotta get back now! And really sorry that i cannot post pictures or videos of the quad right now. I will surely try to upload once we are done with this project here.