Downloads · 30 days
0
robotwin2-checkpoints
Avakn/robotwin2-checkpoints
robotwin2-checkpoints is a machine learning model from Avakn. Use it for the machine learning task on the model card, and read the license before you ship it in a product.
ACT, and pi0.5 single-task finetuning using B200 GPU on RoboTwin2.0 dataset.
Access
Public
Updated Feb 23, 2026
Repo size
44.6 GB
Likes
0
Public
Repo makeup
Click a slice to open those files.
Other30.2 GB · 68%
At a glance
- Access
- Public
- Created
- Feb 23, 2026
- Updated
- Feb 23, 2026
- SHA
- 338ca908
Model card
From the Hugging Face model README
RoboTwin2 Checkpoints
ACT, and pi0.5 single-task finetuning using B200 GPU on RoboTwin2.0 dataset.
The policies were trained on the following Tasks:
place_phone_standplace_a2b_leftmove_can_pothandover_blockput_bottles_dustbin
Data
- Demonstrations: 50
demo_cleanepisodes per task - Embodiment: aloha-agilex (dual-arm)
- Action dim: 14 (6 DOF × 2 arms + 2 grippers)
- Cameras:
cam_high,cam_right_wrist,cam_left_wrist
ACT (Action Chunking Transformers)
Architecture
| Param | Value |
|---|---|
| Backbone | ResNet-18 |
| Hidden dim | 512 |
| Feedforward dim | 3200 |
| Attention heads | 8 |
| Encoder layers | 4 |
| Decoder layers | 7 |
| Chunk size | 50 |
| KL weight | 10 |
| Action dim | 14 |
| Dropout | 0.1 |
| Parameters | ~83.9M |
Training
| Param | Value |
|---|---|
| Batch size | 8 |
| Epochs | 6000 |
| Learning rate | 1e-5 |
| LR backbone | 1e-5 |
| Weight decay | 1e-4 |
| Optimizer | AdamW |
| Save freq | every 2000 epochs |
Checkpoints
| Path | Seed | Val Loss |
|---|---|---|
ACT/act-place_phone_stand/demo_clean-50/ | 0 | — |
ACT/act-place_phone_stand-run2/demo_clean-50/ | 1 | 0.038 |
ACT/act-place_a2b_left/demo_clean-50/ | 0 | — |
ACT/act-place_a2b_left-run2/demo_clean-50/ | 1 | 0.059 |
ACT/act-move_can_pot/demo_clean-50/ | 0 | — |
ACT/act-move_can_pot-run2/demo_clean-50/ | 1 | 0.036 |
ACT/act-handover_block-run2/demo_clean-50/ | 1 | 0.030 |
ACT/act-put_bottles_dustbin-run2/demo_clean-50/ | 1 | 0.032 |
Each checkpoint directory contains:
policy_best.ckpt— best validation loss checkpointpolicy_last.ckpt— final epoch checkpointpolicy_epoch_{2000,4000,5000,6000}_seed_{0,1}.ckpt— intermediate checkpointsdataset_stats.pkl— normalization statistics
Pi0.5 LoRA (place_phone_stand only)
Fine-tuned from gs://openpi-assets/checkpoints/pi05_base/params using the openpi framework.
Architecture
| Param | Value |
|---|---|
| Base model | Pi0.5 (3B params) |
| PaliGemma variant | gemma_2b_lora |
| Action expert variant | gemma_300m_lora |
| Fine-tuning method | LoRA |
Training
| Param | Value |
|---|---|
| Batch size | 32 |
| Total steps | 20,000 (trained to 9,000) |
| Save interval | 200 steps |
| XLA memory fraction | 0.45 (64 GB pool on H200) |
| GPU | NVIDIA H200 (143 GB VRAM) |
Checkpoints
| Path | Step |
|---|---|
pi05_lora/place_phone_stand/step_5000/ | 5,000 |
pi05_lora/place_phone_stand/step_9000/ | 9,000 |
Environment
- Framework: RoboTwin2.0
- Simulator: SAPIEN with Vulkan rendering
- GPU: NVIDIA H200 SXM (143 GB VRAM)
- CUDA: 12.8
- Created
- Feb 23, 2026
- Updated
- Feb 23, 2026