Ask HN: How to build a Robotics portfolio? | Hacker News Reader

Ask HN: How to build a Robotics portfolio?

I know this is a very broad topic so I'll put details here.

I'm a Software Engineer by profession (with ~5 years of industry experience). What particularly interests me in Robotics is Mechanical Design, Motion Planning and Controls.

I have done a few hobby Computer Vision projects so I could definitely make use of that knowledge as well.

I'm confused about what I should be doing to build a Robotics portfolio. Here are some of the options:

1. Build an actual robot to understand how different pieces (mechanical design, navigation, motion planning, controls) fit together - quadruped, biped, hexapod, quadrotor, toy car?

2. Pick existing work and only work on a specific problem - motion planning for quadruped, obstacle avoidance for quadcopter?

3. Build a robot only in simulation without spending time on building and iterating on a physical robot.

View on HN

For DIY projects and mechatronics modding you can surely refer to the latest, just released “Robot Builder’s Bonanza” 5th ed. book, here from Amazon https://www.amazon.com/Robot-Builders-Bonanza-Gordon-Mccomb/... . More in general, it depends on what you do want to achieve: reinforcement learning is the latest hottie with small robotics and a very good academic lab with a lot of public materials is in Zurich at ETH http://rpg.ifi.uzh.ch

Thanks for the links.

Robotics is really broad field, and just by being a software engineer and also with your computer vision projects you've likely already started building a portfolio that a roboticist will be impressed by.

I think that building a physical robot is extremely valuable to anyone who wants to get into robotics. There's so many problems to solve and things to learn that just won't come up in simulation. That being said it might be better to first build your robot in simulation first, which is also extremely valuable to be able to do.

So, if you want my suggestion I would try and do all three of the things you listed. Pick a problem that sounds easy to start. And try to build a robot in simulation to do just that. Having this simulated environment set up will be invaluable for future endeavors too. Once you finish with that I think it will be extremely valuable to actually build your robot and try out the software there. Almost certainly it won't work the first time but that will be a valuable lesson too.

Another note, walking robots are in general very difficult to do well, quadcopters are more straightforward but still much more involved than the car to build from scratch. The car will be easier to get going, and will be more forgiving mechanically and with the controls. But that means you can quickly get into doing the higher-level motion planning if that's what you're interested in, and there are still interesting control and estimation problems to do there. And believe me even though it sounds very easy you will run into a lot of challenges you probably are not expecting.

Reply on news.ycombinator.com