[IDEA] Simulating sensor data

[xjjak]

Describe the vision. What is to be achieved? What does the end-result look like? Is it a specific feature?

Develop a simulation to simulate data gathered by the IMUs on all fingers for strings (?) that infer a specific keyboard input sequence.

1. Set the position of the sensor on a finger and get the expected values for the sensor's output according to the laws of physics (Single finger with 3, or 6 for index fingers, positions).
2. Then there could also be a mode which purposefully introduces errors and can generate realistic datasets for training purposes (all fingers).
3. Graphical representation, in Unity for example??

Share possible solutions or ideas and approaches.

Just a few thoughts so far:

• configuration of the finger: length between joints
• 3 different simulations for the finger types
  • thumb (2 positions, lateral motion)
  • index finger (6 positions, lateral and frontal)
  • all others (3 positions, frontal motion)
• use inverse kinematics to determine the expected angle
• record typing to determine velocity/acceleration

Why? ...

Additional context Add any other context or screenshots about the feature request here.

[CTropp]

Simulating the sensor data is likely to cause additional errors and is therefore not recommended. I'd suggest to generate the data for the machine learning as accurate as possible. Using a simulation to get a huge amount of data might sound promising, but I have doubts in training an AI on simulated data and then try to apply these on the real machine.