This video was helpful in knowing how to implement machine learning in ROS, thanks for that. But is this just for the demonstration of the above mentioned purpose, or is there a reason for using a neural network for obstacle avoidance. That is, do we really need a neural network? Just curious.
Nowadays, almost all the components of a Robot system are run using Machine Learning, like Perception, Navigation, Localization, Grasping and so on. The task in this video is just a toy example to show how Machine Learning can be applied to one such robot system.
Hello! I may have went over it a bit quickly. The overall idea is to convert the problem to a classification problem. We use one hot encoding to do the same. [1 0] implies move forward and [0 1] implies rotate. However the data we collected was of the form: [0.3 0] for forward and [0 0.3] for rotation. We simply converted these 0.3 values to 1. The code in the video is just a general way of doing so. For example, to convert [0.3 0.01] to [1 0] if any noise is also captured. Hope this clears your doubt. Let me know if have nay further questions.
![[LIVE] นาทีนี้ต้องพี่ทนาย | ทนายแก้ว vs ทนายไพศาล | #คุยให้เด็กมันฟัง EP.47 (02/11/67)](http://i.ytimg.com/vi/HmquaO-PL_M/mqdefault.jpg)
This is so helpful. Thanks for sharing your knowledge!
This is so helpful. Thank you for the video.😃
Happy to know you liked it!
This video was helpful in knowing how to implement machine learning in ROS, thanks for that. But is this just for the demonstration of the above mentioned purpose, or is there a reason for using a neural network for obstacle avoidance. That is, do we really need a neural network? Just curious.
Nowadays, almost all the components of a Robot system are run using Machine Learning, like Perception, Navigation, Localization, Grasping and so on. The task in this video is just a toy example to show how Machine Learning can be applied to one such robot system.
@@RoboticswithSakshay okay..makes sense..got it. Thanks.
Hello, I was wondering if the script would only work for a wheeled robot. I am trying to implement this but to a legged robot.
Can you explain little more elaboratively (may be in separate video) on the logic of how the data is classified into 1s & 0s?
Hello!
I may have went over it a bit quickly. The overall idea is to convert the problem to a classification problem. We use one hot encoding to do the same. [1 0] implies move forward and [0 1] implies rotate.
However the data we collected was of the form: [0.3 0] for forward and [0 0.3] for rotation. We simply converted these 0.3 values to 1.
The code in the video is just a general way of doing so. For example, to convert [0.3 0.01] to [1 0] if any noise is also captured.
Hope this clears your doubt. Let me know if have nay further questions.
@@RoboticswithSakshay Got it. Thank you for explaining your logic of labeling the input data to make it ready for NN training.
Must have took a lot of hard work just to make this one tutorial
Yes, the main work was involved in coming up with this idea, and setting the hyperparameters of the Neural Network.