[HS]

That would be really awesome if you could accomplish something like that. I've seen drones that fly and go in the water. I've seen drones that can fly and roll on land. I haven't seen any designs that could do all three though. I have one suggestion; have you thought of experimenting with whegs instead of wheels?

Thanks, I was quite pleased to have figured that one out, at least in theory. Yup, I didn't know what they were called but I've considered the same type of idea as whegs. Maybe I'll reconsider, but these also have to be airfoils/props, so I'll have to find a happy medium. Currently, I think wheels will be more structurally stable and make for a smoother ride; helping with safety and reliability, (which now seems really important because I'm actually making these). Plus these types of wheels might actually decrease the weight requirement, think of a bicycle wheel, it's a very stable construction (like a bridge or something like that), it doesn't require much in the way of materials. I also wanted to test out whether treads would function like golfball dimples, (yeah, it probably won't affect the performance that much but I'm curious to see what the effects will be).

[WriterOfMinds]

Yeah, if you do wheels then the rim can also function as a blade guard. Whegs wouldn't play as well with co-mounted propellers. (And propeller blades themselves are not great load-bearing structures, usually.)

[MagnusWootton]

being fully offroad is very important,   you need to:

    * traverse up and down stairs
    * go through long grass, and other soft body material, like snow.
    * you need to be able to right yourself when you tip over.
    * swimming and flying is really good, if u can fly it gets over alot of problems by itself.
 
you dont want to be stuck in disabled only zones, like cars are!!   

Being powerful enough to fly is just a matter of enough amps,  but I guess further design could drop the requirement,  but if you want little wings/rotors they need to flap/spin furiously! =)

[OllieGee]

Well hello HS my old friend...
This looks like you have clearly put a lot of thought into this project. I haven't delved too deep into your designs yet but I see you mention you are using an arduino board (like a raspberry pi board I'm guessing) for the CPU. Are you planning on using or creating any code for this? I have basic Python skills and would be very keen to see what commands you are thinking of using for its movement. Please do share any videos of anything worth showing...
Well done on this achievement as I say, you have clearly thought about this and I think it could go somewhere at least as AI research algorithms which could be built upon. Kudos to you HS!
   

[HS]

Hello again, OllieGee, glad you're back!  I'll be sure to post an update if I accomplish anything milestone-ish. Yes, I'm considering using an Arduino or similar to test what the robot can do, then once the design is good enough I'll try introducing some AI elements of my own design. The process may take a while since I don't know how to code yet, but I've given myself until the end of the year to make a proof-of-concept prototype, that seems like enough time, and if it isn't, then it'll be an adventure! Thanks for the encouragement.

[OllieGee]

Can’t wait to see any outputs for this HS. I’d also be happy to feedback on any stage gate milestones if you like!

[HS]

Take a look, the hardware is mostly complete!

The main challenge, wobbly joints, convinced me to add cardboard spacers between the servos and their attachments, as well as some magnets to help steady the joints at 0 and 90 degrees.  I also traded the lead acid battery I had previously for some NIMH AA's, which now serve as counterweights for the wheels.

A closer look at the actuator/joint construction. See the metal washers? Those are there to help with the wobbliness, I had to add three cardboard washers around the steel washers to further steady the joint. Now Blinky doesn’t appear to be slouching all the time.

The final iteration of the joint has a fourth servo to control the locking mechanism. There's a magnet behind each battery pack, the fourth servo on each joint rotates it to detach and snap onto the hull magnets.

[infurl]

Beautiful work HS. 

Be careful with those alligator clips. NiMH batteries can supply enough power to start a fire if you get a short circuit. Unless they're just temporary for testing, I'd replace them with a properly insulated plug and socket arrangement.

[MagnusWootton]

Excellent work.    I'm still working on mine!!!

A tip for your neural network-> If you keep storing the accellerations of the robot whilst measuring its current position and velocity, given what random motor movements it was doing, in a big long list  you might be able to train the synapses of your network on them,  but you need a way of training,  I know a way to, if u want some help.

And then u should be able to predict the accellerations due to the current position and velocity given the motor movements.   If u can do that,   then the robot could do a motor search using this physical recording of what happened to it.

Thats how mine is going to work.

There is a video of a guy doing it on youtube, with spectacular results, but I forget what the video was called cause it was from a while back.

[MagnusWootton]

Sorry for being a little away from everyone,  but I'm pretty sick,   Ive dropped from 40 cigs a day to 7 cigs a day, but its not enough, I have to just quit now.    So im very forgetful and im tired everyday,  i'm still working on my bot tho.

I like the cranks on the wheels of this one!  thats being original.   Can those servos actually get it going along?  the little blue ones aren't very strong.    So its supposed to convert from car into quadcopter? 

I think your neural network is a great idea for doing what I said,  because u should be able to at least get 1 meg samples out of that thing a frame, with an average 555 timer spitting out results out of it,  and that would be perfect with the prediction of accelleration from position and velocity,    but its important for it to be analogue not digital, cause you want to be able to tween the learning set thats going into it to fill in the gaps between the i/o samples its recorded.

Your net will work,  well, I know a way it could work,  if I had to suddenly take over the project.

It being a true physical network has a really cool advantage, that running them on a computer doesnt have,  and its how much output throughput it has,  alot more.

So if u need to learn something like backpropagation,   I think I have a way of doing it,  to pack as much knowledge as u can, into the synapse connections to represent a transform from i to o.

[HS]

No worries, please choose what's best for you, I'm glad you're taking care of your health.

Thanks for the feedback and suggestions! I'm planning to revisit the neural net later but for now I just want to test how well the components hold up while driving around. So I'm planning out a simple arduino program to test that.

The next step will be to create a digital model for 3d printing, then use that more weatherproof incarnation of Blinky to work out the details of flying and swimming. This might be a good goal for the end of next year! Once the hardware is capable I'll think about how to continue developing the AI.

These servos were the only ones cheap enough since this build needs 18, not even counting all the ones where the gears started slipping. They weren't strong enough to lift the wheels before so I added the counterweights and will test them again once the parallel servos have been carefully aligned.

The final iteration of this idea is a quadriphibious quadruped - ie, rover, quadcopter, boat, and submarine, who can be a general helper and coinhabitant. A part of the team.

Here's my initial logic plan for indoor navigation, though further updates may have to wait a bit until I resurrect my PC, which decided to give up now of all times.



It'll be great to see an update on your bot when you're ready.

[MagnusWootton]

Those little flow charts make good a.i.'s for robots, I remember getting one done in year 6, in Primary school when I was 11,  I figure they work in a real robot too, just need to detect where you are in the chart successfully then it can be followed by the computer.

[HS]

I’ve probably taken the Blinky 0.1 design as far as it’ll go. My favorite feature (the prop/wheel articulation) caused most of the problems. Just driving forward often led to an undignified collapse in a tangled heap. Why? Well, even when I did my best to fine-tune the angle of the wheels, slight discrepancies made the wheels not want to roll parallel to each other, especially on carpets. I tried to salvage some basic functionality by fixing the wheel articulation mechanisms in place. This allowed Blinky to drive forward and backward ok, but the necessary new turning method (left-side wheels and right-side wheels rotating in opposite directions) caused the motors to stall. Here I set the motor power to near the minimum required to generate motion, but the geometry created the same result even with additional power.



I did get some of the charging and navigation logic working: eg



I’ll still use this platform to improve the code, as there seem to be a bunch of glitches that appear to multiply if I try to get rid of any of them, plus I’d like to invent some niftier logic. But now creating Blinky 0.2 will be the main focus.

[WriterOfMinds]

Sometimes you just have to try things and find out.

Do you think you'll lose the propeller/wheel articulation, or try some other method of keeping the wheels aligned? I could see some kind of closed-loop control possibly being good enough.

[HS]

Yeah, even though it didn’t work as well as I had hoped, I certainly learned a lot. Not to mention, version 0.1 is already one of my favorite things I’ve made.

I’m definitely keeping the articulation! The current plan is to try stronger servos, a (yup good idea) closed-loop control system, mechanical position locks, plus a smaller 3d printed frame. As far as I can tell, I’ll have to look for some affordable components, get the measurements, get access to AutoCAD or Inventor or learn a new program, design Blinky 0.2 based on the measurements, then outsource the manufacture or save up for a printer.