Learning Fine-Grained Bimanual Manipulation with Low-Cost Hardware
Paper • 2304.13705 • Published • 7
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. This repository contains pre-exported model files optimized for Qualcomm® devices. You can use the Qualcomm® AI Hub Models library to export with custom configurations. More details on model performance across various devices, can be found here.
Qualcomm AI Hub Models uses Qualcomm AI Hub Workbench to compile, profile, and evaluate this model. Sign up 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 1.25.0 | Download |
| QNN_DLC | float | Universal | QAIRT 2.45 | Download |
| TFLITE | float | Universal | QAIRT 2.45 | Download |
For more device-specific assets and performance metrics, visit ACT on Qualcomm® AI Hub.
Option 2: Export with Custom Configurations
Use the Qualcomm® AI Hub Models 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 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 | 6.721 ms | 209 - 209 MB | NPU |
| ACT | ONNX | float | Snapdragon® X Elite | 12.758 ms | 149 - 149 MB | NPU |
| ACT | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 8.59 ms | 4 - 501 MB | NPU |
| ACT | ONNX | float | Snapdragon® 8 Gen 1 Mobile | 19.508 ms | 4 - 309 MB | NPU |
| ACT | ONNX | float | Qualcomm® QCS8550 (Proxy) | 12.422 ms | 0 - 80 MB | NPU |
| ACT | ONNX | float | Qualcomm® QCS8450 | 19.508 ms | 4 - 309 MB | NPU |
| ACT | ONNX | float | Snapdragon® 8 Elite Mobile | 6.84 ms | 0 - 412 MB | NPU |
| ACT | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 5.977 ms | 2 - 442 MB | NPU |
| ACT | ONNX | float | Qualcomm® QCS9075 | 20.199 ms | 4 - 49 MB | NPU |
| ACT | ONNX | float | Qualcomm® QCS8750 | 6.84 ms | 0 - 412 MB | NPU |
| ACT | ONNX | float | Qualcomm® QCS7181 | 12.758 ms | 149 - 149 MB | NPU |
| ACT | QNN_DLC | float | Snapdragon® X2 Elite | 4.896 ms | 4 - 4 MB | NPU |
| ACT | QNN_DLC | float | Snapdragon® X Elite | 9.178 ms | 4 - 4 MB | NPU |
| ACT | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 6.066 ms | 0 - 338 MB | NPU |
| ACT | QNN_DLC | float | Snapdragon® 8 Gen 1 Mobile | 16.407 ms | 4 - 253 MB | NPU |
| ACT | QNN_DLC | float | Qualcomm® QCS8275 | 44.737 ms | 0 - 284 MB | NPU |
| ACT | QNN_DLC | float | Qualcomm® QCS8550 (Proxy) | 8.548 ms | 4 - 26 MB | NPU |
| ACT | QNN_DLC | float | Qualcomm® QCS8450 | 16.407 ms | 4 - 253 MB | NPU |
| ACT | QNN_DLC | float | Snapdragon® 8 Elite Mobile | 4.696 ms | 4 - 272 MB | NPU |
| ACT | QNN_DLC | float | Qualcomm® SA8295P | 15.267 ms | 2 - 206 MB | NPU |
| ACT | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 3.751 ms | 3 - 299 MB | NPU |
| ACT | QNN_DLC | float | Qualcomm® SA7255P | 44.737 ms | 0 - 284 MB | NPU |
| ACT | QNN_DLC | float | Qualcomm® QCS9075 | 15.789 ms | 4 - 9 MB | NPU |
| ACT | QNN_DLC | float | Qualcomm® QCS8750 | 4.696 ms | 4 - 272 MB | NPU |
| ACT | QNN_DLC | float | Qualcomm® QCS7181 | 9.178 ms | 4 - 4 MB | NPU |
| ACT | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 6.106 ms | 0 - 415 MB | NPU |
| ACT | TFLITE | float | Snapdragon® 8 Gen 1 Mobile | 16.399 ms | 0 - 271 MB | NPU |
| ACT | TFLITE | float | Qualcomm® QCS8275 | 44.759 ms | 0 - 330 MB | NPU |
| ACT | TFLITE | float | Qualcomm® QCS8550 (Proxy) | 8.581 ms | 0 - 4 MB | NPU |
| ACT | TFLITE | float | Qualcomm® SA8775P | 323.031 ms | 3 - 13 MB | CPU |
| ACT | TFLITE | float | Qualcomm® SA8650P | 323.031 ms | 3 - 13 MB | CPU |
| ACT | TFLITE | float | Qualcomm® SA8255P | 323.031 ms | 3 - 13 MB | CPU |
| ACT | TFLITE | float | Qualcomm® QCS8450 | 16.399 ms | 0 - 271 MB | NPU |
| ACT | TFLITE | float | Snapdragon® 8 Elite Mobile | 4.728 ms | 0 - 284 MB | NPU |
| ACT | TFLITE | float | Qualcomm® SA8295P | 15.759 ms | 0 - 209 MB | NPU |
| ACT | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 3.754 ms | 0 - 324 MB | NPU |
| ACT | TFLITE | float | Qualcomm® SA7255P | 44.759 ms | 0 - 330 MB | NPU |
| ACT | TFLITE | float | Qualcomm® QCS9075 | 15.175 ms | 0 - 71 MB | NPU |
| ACT | TFLITE | float | Qualcomm® QCS8750 | 4.728 ms | 0 - 284 MB | NPU |
License
- The license for the original implementation of ACT can be found here.
References
Community
- Join our AI Hub Slack community to collaborate, post questions and learn more about on-device AI.
- For questions or feedback please reach out to us.