One feature of VirtualGrasp is the ability to pre-record a interaction sequence, and later replay the whole sequence or individual interaction segment on specific objects. This is enabled by the sensor record and replay functionality of VirtualGrasp.
With this functionality, you can start the recording at any moment in runtime interaction inside a VR application . The sensor recorder will record the sensor data frame by frame, and also automatically segment the frames into segments (we call interaction segment) according to which object the hand is interacting with.
SensorDB has 2 hand interaction(s): humanoid_handLeft1 has 10 interaction segments(s): segment #0 has 462/462 frames/states segment #1 has 934/934 frames/states, on object baseball segment #2 has 169/169 frames/states segment #3 has 568/568 frames/states, on object apple segment #4 has 178/178 frames/states segment #5 has 631/631 frames/states, on object baseball segment #6 has 152/152 frames/states segment #7 has 593/593 frames/states, on object pear segment #8 has 40/40 frames/states segment #9 has 358/358 frames/states, on object screwdriver humanoid_handRight1 has 1 interaction segment(s): segment #0 has 4085/4085 frames/states
Above code block shows a display of one recording on an avatar with a pair of left and right humanoid hands. You can see in this recording:
- the left hand sensor data has been segmented (automatically during recording) into 10 interaction segment, with incremental ids starting from #0.
- segment #0 has 462 frames while hand was not interacting with any object (hence no information of on which object),
- segment #1 has 934 frames while hand was interacting with baseball,
- segment #2 starts when hand stopped interaction with baseball, and back to a state without interacting with any object,
- segment #3 starts when hand started to interact with another object apple, and so on …
- however for the right hand, during the entire time of sensor recording (4085 frames), was not interacting with any object, therefore no separated segments.
Why Segmenting into Interaction Segments?
The reason we segment the sensor data into different interaction segments by objects is because we can later replay the sensor data
- either on the entire recording sequence,
- or play just one particular interaction segment on a particular object.
Each option has its use cases. For example once you record all the steps to go through a training task,
- if you want to show the entire steps you can fully replay the whole sequence;
- you can however also replay a specific interaction segment on a specific object to show a single step in training.
What is Recorded Exactly?
What is recorded is the sensor data – wrist position, orientation and grasp trigger signals. In other words, what we record is user provided control signals on hands – where a user places hand, and when a user want to grasp and interact with an object. And what we DO NOT record is the “result” of these control signals, namely, when and how a hand formed the grasp configuration.
As a result, for example, if you recorded a set of sensor data on a particular scene with a set of objects placed at specific locations, and later you tried to play this sensor data on a different scene with either same set of objects placed at different places, or with different set of objects, the resulting interaction will NOT be the same.