Training data from simulation and physical experiments

Description:
This dataset includes the training data and trained models used in our study on autonomous microrobot control via model-based reinforcement learning (MBRL). The data encompasses both simulation and physical experiment results used to train and evaluate an ultrasound-actuated microrobot navigating microfluidic environments.

Specifically, the repository contains:

• Training logs and reward trajectories for MBRL agents across six environments: Racetrack, SPA, Multi-Output Tributary, Vascular, Mixed, and Randomized channels.
• Trained model checkpoints for each of these environments.
• Fine-tuning results from real-world experiments, including target success rates