2 replies
Regarding your question, the main motivation behind this project can be summarized as follows :
- Having an complete perception and grasping pipeline that relies on learning (more specifically Deep Learning).
- Extending the pose estimation and grasping project done last year to have full control on the robot arm (without using the simulator IK solver).
I guess I understand why you are gravitating towards RL, one more thing I would like to ask is, is this project going to include implementation of a few SOTA algorithms and comparing them in this setting, or its about achieving the best possible results on the arm.
Plus, would you recommend that someone like me take up this project, I recently finished studying Policy Gradient Methods, but am yet to read papers on the famous algorithms like A3C, SAC etc.
Moreover, anyone with some RL knowledge and familiarity with basic Deep Learning should be good to go with this project. However, the more knowledge and understanding of literature you have, the more chance you have of landing a good proposal for this one.
March 29 18:00 UTC Student application period begins
April 13 18:00 UTC Student application deadline
I was following this tutorial (https://github.com/robocomp/robocomp/blob/development/doc/DSR-start.md)
But in the section where we are supposed to navigate to idserver folder and make it (step 3 and step 4), I am getting this error
robocomp-robolab/components? I am a little confused about that since LearnBlock contains learnbot_dsl/components, but previously GSoC students added components in robocomp-robolab/components for eg, emotionrecognition. Any resource regarding LearnBlock will also be helpful.
I have a doubt regarding LearnBlock that does it uses
robocomp-robolab/components? I am a little confused about that since LearnBlock containslearnbot_dsl/components, but previously GSoC students added components inrobocomp-robolab/componentsfor eg, emotionrecognition. Any resource regarding LearnBlock will also be helpful.
The main idea is to put in robocomp-robolab any component that could be useful for more than one project. And keep the components tightly related to a project in their own components folder and repository...
I'm currently considering these two papers :
- https://arxiv.org/abs/2007.04499
- https://arxiv.org/abs/2007.00982
There are also plenty of review papers out there that can help you get started.
However, of course, every candidate is free to propose other solutions of choice, with justification.
We are currently considering open-source datasets. Also, they should be more related to indoor activities, unlike KITTI.
Some options are :
- YCB Videos Dataset : mainly used for pose estimation but contains depth information as well.
- NYU Depth Dataset : a commonly-used dataset for indoor segmentation and depth estimation.
When I run rcis simpleworld.xml command in terminal ....why robot is not moving , it is an static position. Pleae help me to sort out from this.
Thank You.
Thanks bro. Now it recognized but while running the test component it says "Error taking camera feed. Make sure Camerasimple is up and running" but my camerasimple and handGesture component is running parallely in other terminal. Why is that so ?
Hello,
I solved that previous error by setting maximum memory size in ice configuration. Now the test component is working fine and I can see that classes are detected in terminal but I can't see that camera window (the handGestureClient window) which put these bounding boxes on the hand and say which class it belongs to, but in handGesture and HandGestureClient terminal windows I can see the hand got recognized as a statement. So, What to do in order to make handGestureClient window camera to work ?
Thank You