Anti-Collision System for Multirotors

cyril.anthony777

Junior Member
Hi Everyone, This thread is about a programmable anti-crash system on which i have been working on for couple of months and i need your suggestions .

It is capable of doing obstacle avoidance, shortest path measurement, 3D mapping(Kinect or xtion ), object recognition and tracking using SURF and Color tracking.

It is a PCB which comes between your flight controller and your receiver

View attachment 24178
There are Two Different boards.
Beginner and Advanced

Beginner:

lpc.PNG

It has a Phillips LPC2138 working @60Mhz , 6ch input, 5ch output, 5 IR Sensors Header and a USB port.
it uses PID as the control algorithm , tuning and programming can be done via USB.





Advanced:

It has a Quadcore Exynos 4412 Processor , an AtMega2560 , 2GB DDR2 RAM , few USB HOST ports , a USB Device port , few headers for IR sensors , few GPIO ports, 10DOF IMU and a composite AV Out.
It has a Micro SD Slot which supports upto 64GB.


This board is for people who are interested in aerial robotics.
it will be available under USD 120 along with IR Sensors and HD WEBCAM

The Atmega2560 takes care of the collision avoidance with walls and other objects that the IR sensor can pick up.
The Exynos 4412 is for image processing which can be used for tracking objects and recognizing incoming objects and avoiding them. You can use any HD webcam with the board and record video onboard and transmit it via 5.8GHz video tx and no OSD required.

Please let me know if you have any questions or suggestions.

Thanks.
 

Craftydan

Hostage Taker of Quads
Staff member
Moderator
Mentor
Cyril,

Those are impressive claims -- onboard collision avoidance at the speed and weight budgets of a multirotor are hardly trivial.

Do you have any demo videos? Any plans for outdoor sensors (many of us don't have a handy gymnasium to shield IR sensors from the sun).
 

Craftydan

Hostage Taker of Quads
Staff member
Moderator
Mentor
What kind of sense ranges are you expecting? Any design goals around reaction time/speed envelope?

The processor and board specs seem nice, and not impossibly slow for the task . . .

. . . but $120 for what amounts to a development board seems a bit high these days, especially for something unproven even as a dev board. Inexpensive for something with a high-end, full-featured working ROM in a functional state, but you're 2 months away from proving it . . . or not. Plenty of airframes fail to perform at these competitions.

Nothing personal -- I truly wish you good fortune -- but without evidence of success, I think you've jumped the gun on your product release announcement.
 

cyril.anthony777

Junior Member
Sorry for the misunderstanding this product is not going to be released for another 5 to 8 months until it is fully developed. We are currently using radxa Rock board for now and later we will be switching over to the Odroid-u3 due to the size and currently we are seeing around 90ms for reaction from surf detection and tracking . it is quiet enough for tracking the ground robot. For collision avoidance we are not using IR or sonar we have something else very effective . as far as we have come we are sure it is possible. well this thread is only about suggestions and we have a prototype which is equivalent to the above mentioned specs and it is working .
We just wanted to know what can be improved or added im not going give out something unproven . The price is also uncertain
 

Craftydan

Hostage Taker of Quads
Staff member
Moderator
Mentor
ah. ok. I see that now.

Sorry, the first post threw me off -- you ask for help and suggestions, but didn't ask any questions or outline any of your challenges, then you put a price/specs on a board with little detail or proof it's flown. It seemed like a veiled attempt at selling prototypes . . . and a poor one at that.

I expect the IR/Sonar to perform well for a ground robot -- they've been trivial on 8-bit micros a over a decade ago when I was in post-grad. Most machine vision at the time was offloaded to a laptop or similar horsepower processor (which isn't that hard to find these days in a high end embedded processor like your ARM). Even then it was a bit slower than you'd want for a multirotor . . . but time marches on -- it's all in the algorithm . . .
 

joshuabardwell

Senior Member
Mentor
I would LOVE to see a video of this system in action. I have seen similar things, but they all rely on off-board processing, and often also require IR markers on the objects and the copter.
 

RichB

Senior Member
I have to admit I'm impressed by you coming back with a product. The number of new project announcements compared to actual releases is terrible in the multirotor world.

You need a better video with longer range sensing and more impressive avoidance and routing. For example, lets see your quad actually moving towards a wall at speed, and stop itself from hitting.

Also, your ebay site won't let you ship to US and doesn't offer a price in USD
 

Craftydan

Hostage Taker of Quads
Staff member
Moderator
Mentor
I have to admit I'm impressed by you coming back with a product. The number of new project announcements compared to actual releases is terrible in the multirotor world.

You need a better video with longer range sensing and more impressive avoidance and routing. For example, lets see your quad actually moving towards a wall at speed, and stop itself from hitting.

Also, your ebay site won't let you ship to US and doesn't offer a price in USD

I'm surprised too this made a production run . . . especially as a daughter-card to a FCB, and not a FCB itself.

And $44 USD is far better than original $120 . . . but still a VERY pricey add-on for something that trips ~6" away . . . at night. IR sensors just don't like the daylight or range . . . but I've already stated that they have problems.

If it's even traveling a fair speed, I expect the "sense" will be far too late for the "avoid" to matter.

But either way, a cute parlor trick. I'd also like to see it respond to more than one sensor simultaneously (are there blind-spots at the motor booms?) and see how it responds to conflicting commands (flying between obstacles).

I have to say the not even 30s youtube video shows a neat parlor trick, but still not a ready or proven product, IMO.
 
Last edited:

cyril.anthony777

Junior Member
Hey Dan ,
you are right if the quadcopter is travelling really fast the system can't sense the obstacle. This is an experimental board so firmwares and hardware will change with coming revisions . After final revision it will be priced at 15-20USD (ebay has very high tax rate in my country ie 13-20%) few of the people have been testing this board and they came with positive results . Currently there is a temporary ban of multirotors in my city (Sad Story)
im unable to take any videos

here are some data
one software loop takes 2ms but the ppm output is refreshed once every 22ms(problem) the ADC of ATmega chips has very low sample rate and high settling time so i have used AD7998 and transferred the data to atmega328 via i2c at 400khz .Another atmega328 is us used to read RC inputs and transfer them via SPI @ 4MHz to the main chip. If you want to cover the blind spot you can use upto 8 sensors with wider detection range and faster response time.The detection range has been set to 2.5 meters for current tests.
this similar detection speed and range is what i have experienced with UTM-30lx LIDAR . If you are looking for more robust system checkout the hokuyo sensors with atom processors