Sharing the best locomotion models

[LifeIsStrange]

It would be nice if the community could share the SOTA locomotion only or mobile manipulation models.
A new (locomotion only) one published today seems notable as:

1. they test it on LeKiwi (hence should also work for aloha mini)
2. they work zero shot without odometry
3. they don't seem to require LIDAR / ultrasound / TOF sensors
4. it has memory and avoid obstacles in real-time

https://steinate.github.io/logoplanner.github.io/
Though it's still not the graal, I'm looking for models that can identify in zero shot, objects and places based on implicit natural language requests, that scale to kilometers distances and that allow seamless switching to a manipulation model (ideally with great initial pose).

[liyiteng]

Thanks a lot for sharing this — this is exactly the direction I believe low-cost mobile manipulation should move towards.

Vision-only, zero-shot locomotion models that generalize across platforms like LeKiwi and AlohaMini are a key focus for us.

Our goal is to make AlohaMini a reproducible real-world benchmark platform for validating these emerging navigation approaches.