Real-World Side-by-Side Comparisons

Real-World Side-by-Side Comparisons#

This page provides side-by-side comparisons between our UrbanVerse-PPO policy trained in UrbanVerse scenes, and prior state-of-the-art methods (PPO-UrbanSim, CityWalker, S2E, and Nomad) in real-world street environments across two robot platforms.

All policies are deployed in a zero-shot setting and evaluated under identical robot configurations and real-world scenes, with no fine-tuning or additional training on the target environments.

Note

The videos shown below are the recordings of the actual real-world experiment from which the results reported in the paper were obtained.