The models are located in project_root/Data generation
This is the same algorithm as Space explorer with the only difference being that the inverse kinematic function is based on the pre-trained neural network. Then all the space is explored as previously and a new error array is defined.
With this approach, both position and orientation errors are below the requirements. In particular, both X and Y error is below 10^-5 (10 nm) and the orientation error is below 40 nrad. This approach outperforms the error base by at least 2 orders of magnitude on the orientation accuracy.
