---
license: apache-2.0
tags:
- pi0.5
- robotwin
- embodied-ma
---

# embodied_ma

Finetuned pi0.5 20k checkpoints for RoboTwin tasks.

- `dump_bin_bigbin/pi05_20k_left_right_lora`: 20k-step pi0.5 LoRA/safetensors checkpoint trained on the left/right split 100-episode dataset derived from 50 official Aloha-AgileX clean demos.
- `handover_block/pi05_20k_left_right_lora`: 20k-step pi0.5 LoRA/safetensors checkpoint trained on the left/right split 100-episode dataset derived from 50 official Aloha-AgileX clean demos.

Each checkpoint folder includes `model.safetensors` and the corresponding `assets/.../norm_stats.json`.