README.md

---
library_name: pytorch
license: other
tags:
- android
pipeline_tag: robotics

---

![](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/act/web-assets/model_demo.png)

# ACT: Optimized for Qualcomm Devices

ACT (Action Chunking with Transformers) is a robotic policy model that is trained to predict the next chunk of actions that the robotic hand is expected to perform.

This is based on the implementation of ACT found [here](https://github.com/tonyzhaozh/act).
This repository contains pre-exported model files optimized for Qualcomm® devices. You can use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/v0.59.0/src/qai_hub_models/models/act) library to export with custom configurations. More details on model performance across various devices, can be found [here](#performance-summary).

Qualcomm AI Hub Models uses [Qualcomm AI Hub Workbench](https://workbench.aihub.qualcomm.com) to compile, profile, and evaluate this model. [Sign up](https://myaccount.qualcomm.com/signup) to run these models on a hosted Qualcomm® device.

## Getting Started
There are two ways to deploy this model on your device:

### Option 1: Download Pre-Exported Models

Below are pre-exported model assets ready for deployment.

| Runtime | Precision | Chipset | SDK Versions | Download |
|---|---|---|---|---|
| ONNX | float | Universal | QAIRT 2.45, ONNX Runtime 1.27.1 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/act/releases/v0.59.0/act-onnx-float.zip)
| QNN_DLC | float | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/act/releases/v0.59.0/act-qnn_dlc-float.zip)
| TFLITE | float | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/act/releases/v0.59.0/act-tflite-float.zip)

For more device-specific assets and performance metrics, visit **[ACT on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/act)**.


### Option 2: Export with Custom Configurations

Use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/v0.59.0/src/qai_hub_models/models/act) Python library to compile and export the model with your own:
- Custom weights (e.g., fine-tuned checkpoints)
- Custom input shapes
- Target device and runtime configurations

This option is ideal if you need to customize the model beyond the default configuration provided here.

See our repository for [ACT on GitHub](https://github.com/qualcomm/ai-hub-models/blob/v0.59.0/src/qai_hub_models/models/act) for usage instructions.

## Model Details

**Model Type:** Model_use_case.robotics

**Model Stats:**
- Model checkpoint: act
- Input resolution: 480x640
- Number of parameters: 83.92M
- Model size (float): 255M

## Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
|---|---|---|---|---|---|---
| ACT | ONNX | float | Snapdragon® X2 Elite | 6.763 ms | 4 - 4 MB | NPU
| ACT | ONNX | float | Snapdragon® X Elite | 12.843 ms | 64 - 64 MB | NPU
| ACT | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 8.618 ms | 4 - 514 MB | NPU
| ACT | ONNX | float | Snapdragon® 8 Gen 1 Mobile | 19.378 ms | 1 - 317 MB | NPU
| ACT | ONNX | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 12.237 ms | 0 - 80 MB | NPU
| ACT | ONNX | float | Qualcomm® QCS8450 | 19.378 ms | 1 - 317 MB | NPU
| ACT | ONNX | float | Qualcomm® Dragonwing™ IQ-9075 | 17.478 ms | 4 - 10 MB | NPU
| ACT | ONNX | float | Qualcomm® Dragonwing™ IQ-X7181 | 12.843 ms | 64 - 64 MB | NPU
| ACT | ONNX | float | Qualcomm® Dragonwing™ Q-8750 | 6.868 ms | 0 - 412 MB | NPU
| ACT | ONNX | float | Snapdragon® 8 Elite Mobile | 6.868 ms | 0 - 412 MB | NPU
| ACT | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 5.965 ms | 1 - 446 MB | NPU
| ACT | QNN_DLC | float | Snapdragon® X2 Elite | 4.807 ms | 4 - 4 MB | NPU
| ACT | QNN_DLC | float | Snapdragon® X Elite | 9.208 ms | 4 - 4 MB | NPU
| ACT | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 6.079 ms | 2 - 338 MB | NPU
| ACT | QNN_DLC | float | Snapdragon® 8 Gen 1 Mobile | 16.458 ms | 4 - 254 MB | NPU
| ACT | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8275 | 44.715 ms | 1 - 286 MB | NPU
| ACT | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 8.516 ms | 4 - 5 MB | NPU
| ACT | QNN_DLC | float | Qualcomm® SA8775P | 13.808 ms | 1 - 269 MB | NPU
| ACT | QNN_DLC | float | Qualcomm® SA8650P | 13.808 ms | 1 - 269 MB | NPU
| ACT | QNN_DLC | float | Qualcomm® SA8255P | 13.808 ms | 1 - 269 MB | NPU
| ACT | QNN_DLC | float | Qualcomm® QCS8450 | 16.458 ms | 4 - 254 MB | NPU
| ACT | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-9075 | 14.855 ms | 4 - 9 MB | NPU
| ACT | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-X7181 | 9.208 ms | 4 - 4 MB | NPU
| ACT | QNN_DLC | float | Qualcomm® Dragonwing™ Q-8750 | 4.695 ms | 0 - 272 MB | NPU
| ACT | QNN_DLC | float | Qualcomm® SA7255P | 44.715 ms | 1 - 286 MB | NPU
| ACT | QNN_DLC | float | Qualcomm® SA8295P | 15.253 ms | 0 - 205 MB | NPU
| ACT | QNN_DLC | float | Snapdragon® 8 Elite Mobile | 4.695 ms | 0 - 272 MB | NPU
| ACT | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 3.778 ms | 4 - 299 MB | NPU
| ACT | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 6.113 ms | 0 - 418 MB | NPU
| ACT | TFLITE | float | Snapdragon® 8 Gen 1 Mobile | 16.378 ms | 0 - 271 MB | NPU
| ACT | TFLITE | float | Qualcomm® Dragonwing™ QCS8275 | 44.742 ms | 0 - 325 MB | NPU
| ACT | TFLITE | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 8.496 ms | 0 - 3 MB | NPU
| ACT | TFLITE | float | Qualcomm® SA8775P | 13.821 ms | 0 - 280 MB | NPU
| ACT | TFLITE | float | Qualcomm® SA8650P | 13.821 ms | 0 - 280 MB | NPU
| ACT | TFLITE | float | Qualcomm® SA8255P | 13.821 ms | 0 - 280 MB | NPU
| ACT | TFLITE | float | Qualcomm® QCS8450 | 16.378 ms | 0 - 271 MB | NPU
| ACT | TFLITE | float | Qualcomm® Dragonwing™ IQ-9075 | 14.061 ms | 0 - 71 MB | NPU
| ACT | TFLITE | float | Qualcomm® Dragonwing™ Q-8750 | 4.74 ms | 0 - 284 MB | NPU
| ACT | TFLITE | float | Qualcomm® SA7255P | 44.742 ms | 0 - 325 MB | NPU
| ACT | TFLITE | float | Qualcomm® SA8295P | 15.757 ms | 0 - 207 MB | NPU
| ACT | TFLITE | float | Snapdragon® 8 Elite Mobile | 4.74 ms | 0 - 284 MB | NPU
| ACT | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 3.747 ms | 0 - 325 MB | NPU

## License
* The license for the original implementation of ACT can be found
  [here](https://github.com/tonyzhaozh/act/blob/main/LICENSE).

## References
* [Learning Fine-Grained Bimanual Manipulation with Low-Cost Hardware](https://arxiv.org/abs/2304.13705)
* [Source Model Implementation](https://github.com/tonyzhaozh/act)

## Community
* Join [our AI Hub Slack community](https://aihub.qualcomm.com/community/slack) to collaborate, post questions and learn more about on-device AI.
* For questions or feedback please [reach out to us](mailto:ai-hub-support@qti.qualcomm.com).