You need to agree to share your contact information to access this model

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

pi0.5 S1 Mobile LoRA

Four independently fine-tuned pi0.5 LoRA policies for the S1 Mobile robot.

Task	Config	Checkpoint directory	Training step
trash-garbage	`pi05_s1_trash_garbage_lora`	`trash-garbage/`	19999
trash-bottle	`pi05_s1_trash_bottle_lora`	`trash-bottle/`	19999
trash-can	`pi05_s1_trash_can_lora`	`trash-can/`	18000
pour-blue	`pi05_s1_pour_blue_lora`	`pour-blue/`	19999

Each checkpoint directory contains inference params/ and task-specific assets/. Training-only train_state/ is intentionally excluded.

The complete source tree is under deployment/openpi/. It is based on official OpenPI commit 15a9616a00943ada6c20a0f158e3adb39df2ccac and includes exact files copied from the training server.

Packaging validation completed successfully: the S1 source/config tests passed (7 passed), and the formal trash-garbage checkpoint restored and produced a finite (32, 32) action chunk in the deployment smoke test.

Start with:

Important: the 32D output is a mixed command representation. Torso and arm outputs are absolute end-effector pose targets, grippers are absolute, and chassis indices 29:32 remain (dx, dy, dtheta). All relative training targets use the same observation at the beginning of the 32-step chunk, not adjacent-action rolling deltas.

Downloads last month: -; Downloads are not tracked for this model. How to track

Video Preview

Robotics