How to create real-life robots?

Question

Even before I learnt programming I've been fascinated with how robots could work. Now I know how the underlying programming instructions would be written, but what I don't understand is how those intructions are followed by the robot.

For example, if I wrote this code:

object=Robot.ScanSurroundings(300,400);
if (Objects.isEatable(object))
{
   Robot.moveLeftArm(300,400);
   Robot.pickObject(object);
}

How would this program be followed by the CPU in a way that would make the robot do the physical action of looking to the left, moving his arm, and such? Is it done primarily in binary language/ASM?

Lastly, where would i go if I wanted to learn how to create a robot?

Answer 1

You would have to have a driver that interfaced with the hardware (most likely a STAMP or FPGA with motors etc...). You would then call the function me.moveLeftArm(x,y); and the driver would know that moveLeftArm() means to move an actuator for X seconds/milliseconds/degrees.

I'm sure that you could find a kit that does robot programming.

Answer 2

Is probably easier to get a more high-level language to describe the robot's behaviors and intelligence and let the low level language to the actions (move arm, walk, stop). There is a lot of research in what is called BDI architecture for intelligent agents, google for it.

You can find more about at this site, it's a DSL for describing agent behavior made in Java. It's called Jason interpreter and the language is AgentSpeak(L).

Answer 3

You should check out Microsoft Robotics Studio (MRS). They have many videos/screencasts, and written tutorials. Additionally, Channel9 has many videos, interviews, etc, on the robitics subject. Including demonstrations, and interviews with developers of MRS.

Answer 4

Robots will work by interacting with hardware. The bridge from your code is often done through different type of I/O ports. It could simply be a RS232 cable for example (you know those old COM1 ports). Hardware parts will be composed by motors (such as servo motors) and sensors (such as ultrasound to feel obstacles, lasers to get distance or switches).

You don't need to use assembler to do that, there are lots of languages (if not most) that can do it but it requires knowledge on how to interact with hardware. Like writing a driver. It requires at least basic electronics also if you want to build the robot yourself.

If you're interested, I suggest you have a look at this book which is a good primer.

Also, you could try out programming a Basic stamp, it's pretty easy following the tutorials and it will give you a good start on how to build robots. It's not too expensive and you'll be interacting with hardware in no time.

Good luck and have fun!

Answer 5

If you get good enough at programming, you may discover that you don't even actually need a robot to test much of the hardest code you'll need to write... (IE, making a robot see and recognize a scene always fascinated me... But at some point, I realized that the physical robot required for this problem is the easy part... The software is the hard part!)...

Answer 6

In the end, something has to break down the high level commands into very low level commands. Something has to translate "Pick up the cup" to how to move the arm (what angles the joints should be at) to the hardware commands which actually turn the motors.

There are frameworks which try to provide some amount of this translation, including (but not limited to):

• Player/Stage
• Microsoft Robotics Studio
• Carmen
• CLARAty
• Lego Mindstorms

However, since robotics research is interested in every layer of the system, there aren't many systems which provide the entire translation stack. If you're looking into getting into robotics, there are several systems which attempt to make this easier (again, a random sample):

• Lego Mindstorms
• TeRK
• VEX Robotics

Failing that, sites such as Make even provide guides to building robot projects to start from. The challenge is find a project which you are excited about, and go to town!

Answer 7

In most modern robots you would have an Inverse Kinematic model of the mechanism, in this case the arm, that converts the spatial coordinates into positions for the joints of the arm. These joints are usually moved by servo motors. To smoothly move the arm, you need a series of intermediate joint positions defining the path you want the arm to follow. You also have to worry about the velocities of the joints, which together control the speed of the "hand" at the end of the arm.

While the arm is moving your servo system will be getting feedback about its actual position. Simple servo systems may use a basic PID feedback loop to adjust the motors. More complex systems will include feed-forward parameters which compensate for inertia, gravity, friction, and so on. These can become very sophisticated.

The real fun starts when you have to allow for obstacles in the space around the robot. You have to sense the obstacle and figure out how to avoid it and still reach the destination.

Answer 8

If you want a Java alternative, I can recommend the book Linux Robotics. It has a lot of good information about where to get kits, parts, and sensors, as well as complete source code listings in Java.

Answer 9

Find a local FIRST robotics team and volunteer to be a mentor. FIRST is a robotics competition for middle and high school kids. The goal is that the kids do all of the work to build, program, test, and run the robot, but you still will have lots of opportunities to dig in and really learn the software. They are using LabView by National Instruments, and, as of Feb 8, have just begun regional competition for this year. LabView is a graphical programming environment that interfaces with NI hardware to let you program motors, actuators, and sensors. The NI stuff is pretty slick and is pretty easy to use, plus it's provided free to each team, so you don't have to buy the hardware and software yourself (at least to get started.) Plus, you get the added bonus of helping a new generation of engineers get their start.

Answer 10

I share the same itch .. I'm about to buy my first Beagle Board and some sensors / servos that can use the I2C bus. I'm going to be using an event driven design and a crude implementation of fibers (fibrils, if you will) which are userspace threads.

Basically my design calls for one process, which launches one thread per group of servos. Each group manager thread will launch x # of fibrils, 2 per servo (likely). One fibril is used to control the servo, the other fibril handles events from that servo (i.e., an object is just too heavy to pick up, an object was dropped, etc).

The main process has the task of listening for events from everything else and making sure the 'right hand knows what the left hand is doing' while moving forward and negotiating obstacles.

Its going to take me the better part of two years to get something working to the point that I'm proud of it .. but I anticipate many enjoyable evenings getting it to that point.

I will very likely be using a Microkernel, not Linux.

I'm doing this as much to sharpen myself with event driven methods as well as my desire to make my own R2 :)

Answer 11

Start with Phidets if you are familiar with .Net. You can checkout TrossenRobotics.com for parts.

The Phidgets interface kit is a good place to start. From there you can get a servo controller and start building things that move.

The Trossen forums are also a good place to review other people's projects. They have a new Data Center with code/project samples too. I don't work for them...just a happy customer.

Answer 12

lots of good answers here. your piece of fantasy code is not far from how you'd do in a higher level language such as C# over MS Robotics Studio. Just keep in mind that even simple things (like "move arm left") are very loaded with "information bias".

down to the metal, a robotic arm is a set of links and [possibly] motorized joints. Therefore "move arm left" (or any point in coordinate) is already a very complex task to compute (look for D&H Table, forward and inverse kinematics for manipulators).

There's also the concept that move arm left assumes there's nothing in that space and a collision won't occur. If the environment is unconstrained, then you need to implement a collision detection system, often based on some sort of sensor (camera) and machine vision algorithms.

In summary, the language and the hardware interfacing are often trivial compared to modeling the system to achieve the desired behavior.

Answer 13

I just have to add something about Arduino projects to this because I dont see it mentioned above.

There is a very low bar for entry into the Arduino based robotics projects. The "sketch" programs that you write for the hardware are very easy to pick up and similar to C syntax. If you dont know your transistors from resistors these boards still allow you to do alot with plug-in hardware and additional "shields" that extend the base computer board.

Its very fun, very flexible and something to get your code interacting with the real world. Plus its "Open Hardware" very along the lines of open source software.