Update 46-3580.mp4. We use additional 10h to train ACT on real dataset, and it works!

(Updated Version) We built the simulation environment and digital twin for the dual-arm SO100 from scratch and collected data in the Maniskill virtual environment. We used ACT to train and verify a dual-arm collaborative pouring task. We further collected data in the real-world environment and fine-tuned the ACT model for final validation.

46-3580.mp4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1762665a149b7772d1b8856b20f0d6de5392305c063534e9504a913c479e588e
-size 107250190
+oid sha256:fb9608d2ab130be668d95bfd6ae4c350fcac63b17036fdd3b02b43ed75fba938
+size 135216258