Introduction to Autonomous Robots | Hacker News Reader

Introduction to Autonomous Robots

github.com

Shameless plug, but I'm working with a startup that teaches middle-school (and up) autonomous robot programming within the context of Sumo Robot competitions. We're active in 6 schools right now in Georgia and hope to expand to the entire east coast this year.

What sets our program apart is that we've developed a super low-cost robotics kit that lets us sell each student their own programmable robot for $100 each. I just spent the last four months writing the 8-week curriculum and associated text book.

Our website is at http://www.sumorobotleague.com/ and we'd love for you to sign up for our mailing list. We're trying to figure out manufacturing right now so that we can crank out thousands of robots to keep up with demand. If you're interested in robotics, I think our kit and program is probably the easiest introduction that you'll find.

Would this be applicable for someone with lots of programming experience, but zero automation / robotics experience?

I'm a hands on, self learner. I don't do well in a structured course, but prefer to learn by tinkering and building (while referencing good resources). I'd love to start with something like this, but wonder since it's marketed to middle school kids, if it would be too trivial?

On the other hand, I have zero experience in this field, so maybe it would be perfect...?

I don't know about this particular robot, but if you're up for it, you can built your own quite easily.

https://dobots.nl/2012/08/26/the-dotty-robot/

https://dobots.nl/2012/06/19/implementing-swarm-behavior-wit...

https://dobots.nl/2014/02/26/dotty-update/

This uses the Arexx frame and on top of that 3D printed hulls (so it's under say around a 30 bucks). We considered Kickstarting it, and we might actually do this one day. Especially with the wireless charging added it to, it's actually quite a cool thing. :-)

Most important of all, this allows you to build normal Android apps to do all kind of interesting experiments. That's more than you can do on any robotic platform!

Yeah, that was kind of the intention. Our kit was designed to be as easy as possible. It just has two motors, a sonar sensor, three IR sensors (for line detection) and a buzzer. The book that I wrote has to take an approach that makes it accessible to kids with no programming experience, but if you're familiar with C++, you'll be kicking ass in no time.

This is the exact sort of setup that I started with (professional engineer of 15 years). Two years ago when we started the program, we built it off of an Arduino and some Pololu kits. It was a great kit, but ended up costing about $150. After the success of our program in the schools, we spent the last year developing a custom PCB that has everything all built in. It has an ATmega328 chip, motor controllers, and sensors all pre-soldered on one board. It's still Arduino programming, but on our custom device. That got our manufacturing costs down to less than $50/kit and lets us sell the whole package to schools for about $100 per kid.

Now the hardest part, as I mentioned, is figuring out manufacturing at scale so that we can crank out thousands of these little suckers.

Send me an email (in my profile) and I'll get you a free copy of the ebook. I'd love some feedback from a programmer, if you're willing.

Right on. I'm definitely going to get one of these then. I'll shoot you an email.

The summary is dense, yet the chapters remain readable. The figures disseminated across the pages make it really easier to grasp the concepts.

I don't know anything about robots, but shuffling thru the pages I found similarities with the video game programming field.

Thank you for this !

I'm in a robotics course right now and the midterm is to build a simulation of a robot with mecanum wheels, where you can provide paths or shapes and it will execute. It sounded daunting until I realized I should just write a game loop and update the robot each tick.

You can do that in simulation fine but when you're doing planning you're eager to speed things up by moving as much into analytic solutions as feasible.

I'm surprised that a chapter titled "Intelligence and embodiment" (ch.1) would not include even a reference to Rod Brooks (MIT Mobile Robot Lab).

Brooks' invention of "subsumption architecture" (augmented finite state machines) and layered control systems in which higher level behaviors suppress low-level behaviors has been demonstrated to lead to robust intelligent behavior without centralized control, internal mapping, and emergent behavior. In short, a biologically-inspired solution which had tremendous influence in the field. To my mind, such an omission is a bit like writing a book on computer graphics and not including a reference to Watt, Foley or vanDam. Not saying it's wrong, just very surprising.

An introductory textbook probably isn't the place to be introducing heterodox views. None of the commercial robotics companies I'm familiar with, including Rethink Robotics where Rodney Brooks works now, use subsumption architectures.

I agree with you. But Roomba.

Right, there is the Roomba. But don't the newer Roombas do mapping?

I think that is true, but as you say that's the most recent model.

The Roomba algorithm is roughly:

- 1. Go in an increasing spiral until you hit something.

- 2. Wall follow for a while.

- 3. Turn away from the wall and go a few meters.

- 4. Go to step 1.

Brooks at MIT, and Latoumbe at Stanford, headed the two opposed camps during the AI Winter. Brooks wanted purely reactive robots, and did some nice insect robots. Latoumbe was into very rigid planning systems, where you approach manipulation as path planning using maze solving algorithms in a N-dimensional space with obstacles. Both turned out to be dead ends. Purely reactive systems don't get beyond insect level, and high-dimensional planning requires total information about the environment.

Then the statistical machine learning people took over and started to get real results, especially in sensor data reduction.

Neither was a dead end. Each made real contributions to the field and have real applications but, not surprisingly, turned out not to be the silver bullet that could do all things. The statistical crowd was always there and have been good at the same set of things, it just finally matured as a field and was able to produce usable results.

They all go through the same motions as anything in this industry. They show promise, everyone gets all worked up about the new thing, it produces some results and gets integrated where it makes sense, gets boring and people start looking for the next thing.

Subsumption was also fast dead end. It didn't scale up. In practice the main ideas were absorbed into the now orthodox two or three layer architecture. IIRC, Arkin was early to show the hybrid reactive+planner approach that we mostly use now.

Can anyone locate the PDF?

https://github.com/correll/Introduction-to-Autonomous-Robots...

thanks ... could not find it either.

Look under the 'Releases' tab. There you'll find different versions of the book, as pdfs

Thanks - the tabs are one level higher, I simply couldn't find it easily, also not easily on mobile.

This is rally cool. Have been looking to get to know about robots a little more. Thank you for sharing this.

For the interested, this is "autonomous" as in locomotion. (Not financially autonomous).

Thanks for sharing

Reply on news.ycombinator.com