1 List of Supplemental Figures
- Fig. S1. The Actor-Critic architecture used to learn optimal control policies for the microrobot.
- Fig. S2. The approximate RBC shape we used to facilitate a fast collision check in the simulation.
- Fig. S3. Scheme of training and evaluation workflow. Efficient training is achieved by selecting increasingly challenging tasks as motivated by curriculum learning.
- Fig. S4. Navigation and localization trajectories of microrobots in free space with different propulsion rotation ratios.
- Fig. S5. Navigation and localization trajectories of microrobots in free space with different external flow fields.
- Fig. S6. Navigation of microrobots in blood vessels with different external flow fields.
2 Supplemental Movies
- Movie S1. Navigation and localization of microrobots in free space with different external flow fields.
- Movie S2. Navigation of microrobots in blood vessels with different vessel diameter D and RBC volume fraction \(f\). Specifically, from left to right, \(D\) = 12um, \(f\) ~ 10%; \(D\) = 25um, \(f\)~10%; \(D\) = 50um, \(f\)~5%; \(D\) = 50um, \(f\)~10%.
- Movie S3. Navigation of microrobots in curved blood vessels with varying cross-section diameter.
- Movie S4. Exhaustive spatial survey in a blood vessel. From left to right, the blood vessel has zero RBCs, ~5% RBCs, and ~10% RBCs.