So, I'm not sure if it's the right place to ask this but it's robotics-related, I use ROS in my job so here you go.

I'm designing a control pipeline and I won't get into the details but there's one section that needs to do the following: A camera should record a person doing a manipulation action/task and a model needs to recognize the action he's doing. I'm not sure if it can be taken as a classification problem where there's a pool of options. I'm thinking the basic manipulation actions like pushing, pulling, pick-and-place, etc. Maybe in the recording, the object that's being manipulated can be detected and the model can recognize from the change of their states thr action that has been done. I'm not sure if I'm clear - a manipulation action is obviously an action that is done by the arm (manipulator).

Are there any papers that tackle this problem/topic that you can recommend?