ROS Resources: Documentation | Support | Discussion Forum | Service Status | Q&A answers.ros.org

[TB3] Additional Hardware for Teaching

I bought some TB3 Waffle Pis with OpenManipulator to supplement a course about autonomous robotics. Which additional hardware (e.g., sensors) would you buy to teach fundamental concepts to university students? My budget is around 3500 EUR.

How much do you have left after the TB3 and arms? Or is that 3500 EUR for just sensing? How many robots do you need to outfit?

What’s the general scope of the course (that would help determine what sensors are just “not in curriculum”.

The curriculum is flexible. In fact I am currently designing the course around the TurtleBot3 and Openmanipulator, so the course will contain theory about robot arm kinematics.

I would like to focus more on the autonomy than control engineering and mechanics, because the course is for computer science students. I am open for other suggestions.

At the end of the course the students should be able to solve something like TB3 home service challenge, or driving autonomously through corridors.

What I stumbled upon is Intel Realsense T265 which seems to even support ROS2, which I want to use in the course. Adding 3D vision would be the next easiest step I thought. The camera costs ~200 EUR.

Another idea was to buy the swarm bundle from Bitcraze to experiment with radio localization, but the bundle is above my budget and I would rather prefer an add-on for TB3.

I also looked at TB3 Friends projects and found the TB3 carrier interesting. The students could maybe build a carrier which could deliver some goods from a room to another, but the project does not include any additional sensors.

Again, I would like to focus more on autonomy. I am open for suggestions :slight_smile:

I bought 4x TB3+Arm already. Now I have around 3500 EUR left (for everything, not only sensors).

Drones are enough for a class by themselves (I literally took a dynamics class in college that went from “here’s a drone” to “its flying autonomously on GPS signals” through kimematics -> dynamics -> control -> optimal control -> collision avoidance -> trajectory tracking. It was a good class.). I wouldn’t try to add in too much especially if you’re trying to cover primarily manipulation.

I think it would be helpful if you provided a budget for just sensors, if that 3500 is also for other stuff.

My recommendations:

  • Astra (https://orbbec3d.com/product-astra-pro/) 4x = ~800 USD. This over Intel depth cameras because intel’s cameras have a bunch of noise and for a class project, I don’t think you want to deal with all of that. Plus TB1 and 2 kits came with Astra cameras so well supported and documented
  • YDlidar X2 or G1 (https://www.robotshop.com/en/ydlidar-x2-360-laser-scanner.html) 5x (one spare) = ~$375 USD. Its more reliable than the RPLidars for a cheaper pricepoint. These are cheap lidars though so expect them to burn out with extended use (maybe 12 months of high usage). Nice thing is they’re cheap so its not a huge deal. 8m would be enough for classroom activities. I’m not sure the pricepoint on the G1, but that will be longer range and much more reliable. I have one though I have some pricier lidars I use for my development platforms.
  • 4 arduinos and 4 BNO-055 dev kits (https://www.adafruit.com/product/4646?gclid=CjwKCAjwmf_4BRABEiwAGhDfSXNlzIkLMRxNaOuvwYKn_T2XpARYxXiv3r9EucBgzQ2Elli8bRcLKBoCNFoQAvD_BwE) = ~200 USD. Good ‘hello world’ examples for ROS using arduino and the IMU is a pretty basic sensor you can design a kinematics lab around. Also good for EKF/KF labs for filtering angular velocities into an orientation estimate. The TB3’s come with IMUs so these would be starters before hitting the robot. Get them a little taste of C and playing with sensor directly (and exposure to I2C).
  • T265’s could be good = ~800 for 4x. They’re good for VIO so I’m not entirely sure what the teaching element here would be other than just something neat. They are decent stereo cameras with hardware synced IMU so if this was a graduate course with visual processing skills, you could have them implement their own VIO or VSLAM. Or use another ROS package to do VIO or VSLAM like ORB-SLAM2 using the images in and just ignoring the VIO on the sensor itself.

That’s about a total of ~1800 EUR.

Now for the fun stuff that you’d have to email some people asking for discounts or just 1 lab piece for use.

  • Ouster OS-1 Lidar ($3500 retail) just 1. This would be good for a 3D perception lab using PCL to do something. These are industrial pieces of equipment used in autonomous vehicles and increasingly on robots. Good labs around floor removal, obstacle detection, really anything with PCL or having them implement some algorithms for depth processing.
  • SICK TiM 571 or a Hokuyo 20-LX (~$2000 retail) just 1. These are industrial lidars that many companies you know use (Fetch, Simbe, Locus, etc)