Robot grab-action-release

Robot:
it doesn’t need that much data to work in realtime,

1. detect 3d surrounding scenery (in realtime), i.e. object space and robot arm state is known and robot physical body is known
2. from task goal : take action A on object/surface U to obtain its future position state
3. identify grab pose, and get robot arm into that pose, knowing the surrounding 3d environment, and avoid robot arm collision or locking with surroundings or itself. Given joint space, and obstruction space, from many possible trajectories (from infinite possibilities if small changes), choose possible. Reiterate algorithm to work on dynamic 3D scene data. As if 4D volume-time data was available, its just one solution , action taken till time t, or updated based on it. that it reaches the final goal
4. action task is complete, wait for next action.