1
$\begingroup$

For a research project at university I have been tasked with simulating a "robot arm". The simulation is to be compared with the real life version for accuracy. The arm will be lifting other objects and building simple structures in the demonstration.

My supervisor has asked that we build this simulator from scratch so I am currently selecting a physics engine.

In this link, it seems bullet is not accurate enough, albeit this was in 2010, for the original poster's needs in robotic simulation.

This comparison believes bullet is better in general but says that documentation is lacking, which is important to me as this is my first time using a physics engine such as this.

so thoughts on these or any other physics engine that may be more suitable?

$\endgroup$
3
  • $\begingroup$ Welcome to Robotics, user15152. As it stands, you don't ask an answerable question. You say, "thoughts on these or any other physics engine that may be more suitable?" Well, what is suitable? Your supervisor wants a simulator built from scratch - Why? What defines "from scratch"? You're in academia - why not use Matlab's physics simulator? You haven't given any scope or criteria for what you're looking for or why, so it's not possible to provide a critique of how one package is more or less suited than another. $\endgroup$ – Chuck Oct 31 '16 at 14:10
  • $\begingroup$ I think, if you had a set of specifications you were looking for, then a question of, "Can software package ___ meet need X?" might be an okay question to ask here, but the people that could answer that question best are the people that made the software package. If you provide specifications and then want to "poll the audience" and get lists of possible software packages, then that question is off-topic and should be asked in Robotics Chat instead. $\endgroup$ – Chuck Oct 31 '16 at 14:14
  • $\begingroup$ So, I'm going to put the question on hold to give you an opportunity to revise your question to include some criteria. What defines scratch? Why re-invent the wheel? What is the goal of the project? Please edit the question to address these points. $\endgroup$ – Chuck Oct 31 '16 at 14:16
1
$\begingroup$

The accuracy of the physics engine will most likely not be a problem. Differences between real world systems and their models are a much larger factor of errors then the slight differences between physics engines so you can use any physics engine you like.

You will not be able to have accurate models (geometry imprecisions, mass characteristic imprecisions, imprecisions in friction constants, etc.)

To my knowledge OpenRAVE can do dynamics and it is meant for robotics, so you will find lots of examples on modelling robots. If you have to start from scratch you can use its API.

$\endgroup$
1
0
$\begingroup$

Isn't using a physics engine cheating if your advisor wants you to start from scratch?

I would choose simbody over those two. It has great documentation and was designed for simulation instead of video games.

What exactly are you trying learn (or predict, or compare) from the simulator? If you want a useful simulation it matters a lot.

The history of robotics is full of simulation projects that died useless because the researcher was not really sure what they were simulating or why.

$\endgroup$
1
  • $\begingroup$ This would depend on what is meant by "from scratch". Using ODE or Bullet directly would be much more involved than just setting up GAZEBO. $\endgroup$ – Jakob Oct 31 '16 at 8:35

Not the answer you're looking for? Browse other questions tagged or ask your own question.