What are the possibilities and limitations of ROS?

asked 2020-05-24 07:48:59 -0600

tigeysen gravatar image

I want to get into Robotics and I have strong background in software engineering/Machine learning. What spiked my interest was a strawberry-picking robot. It seemed so simple, yet it combines so many things that I'm interested in.

Now I've done some research and many people advice to start with ROS. I've skimmed a book quickly and think I get the basic concept.

Now my question is, would it be possible to create a strawberry-picking Robot using ROS and the Gazebo simulator, that would be used in real-life? What would be the path towards this (just as an overview, I'll still first get a good grasp of ROS)?

Things I question:

  1. How would I implement my image recognition models?
  2. How would someone create a simulated version of for example a 'strawberry farm'.
  3. As I understand it, some robotic parts can work with ROS, some cannot, how to know which ones do & don't?
  4. How does this relate to Raspberry or Arduino?
  5. What are good intro sources to get into robotics?
edit retag flag offensive close merge delete


Hi, are you talking about out in the field or in an robot optimized environment?

If optimized: -Navigation on a flat surface, in lanes should be ease. -Identifying ripe strawberrys should be easy by color. -Stereo Camera should work well to aproximate the location of the target -The invers kinematic arm path planning is the computationaly easier the fewer the joints. In this case it might even be possible with just 3. -Raspberry Pi seems to underpowered at first thought. In the end it might be enough but while developing I wouldnt use it. For sure you cant run gazebo on it or rviz. -I havent seen flexible, deformable objects in Gazebo. But if you can pick marbles from some sort of tree like structure it should be enough for simulation as strawberry picking robots use special soft grippers, afaik. So position and grip is the only action needed. -Intro ...(more)

Dragonslayer gravatar image Dragonslayer  ( 2020-05-24 09:13:36 -0600 )edit

part2: intro sources: In the end nothing beats building/experimenting with a real world robot as sensors and real world just dont act as predictable as simulation, and simulation itself is limited. There are still many fields with just unperfect solutions in robotics (no one size fits all). What might work well with gazebo might be chaos IRL. This is also true for the electronics and mechanical parts. Size, weight, power, power consumption, related safety aspects, are all parts that simulation ignores. They can easily be underestimated. And simply adding capabilities might lead to near exponential complex growth of variables, rendering a autonomous mobile robot solution impractical. Thats why I asked for the robots environment, If you build a new glashouse anyway one can make it robotfriendly thus allowing for a simple robot, making it all work. But building a robot that can go out on any field and pick ...(more)

Dragonslayer gravatar image Dragonslayer  ( 2020-05-24 10:07:19 -0600 )edit