Create README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,51 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# Omega Humanoid AI v4
|
| 2 |
+
|
| 3 |
+
## Overview
|
| 4 |
+
Omega Humanoid AI v4 is a high-capacity Temporal Convolution Network (TCN) combined with Transformer Encoder architecture designed for advanced humanoid motion understanding and prediction.
|
| 5 |
+
|
| 6 |
+
## Model Details
|
| 7 |
+
- Architecture: TCN + Transformer Encoder
|
| 8 |
+
- Framework: PyTorch
|
| 9 |
+
- Tasks: Action Classification, Motion Forecasting, Balance Optimization
|
| 10 |
+
- Number of Classes: 12
|
| 11 |
+
|
| 12 |
+
## Input Format
|
| 13 |
+
- Sequence Length: 120 frames
|
| 14 |
+
- Joints: 28
|
| 15 |
+
- Features per Joint: 6 (x, y, z, vx, vy, vz)
|
| 16 |
+
|
| 17 |
+
## Output
|
| 18 |
+
- Action Label
|
| 19 |
+
- Confidence Score
|
| 20 |
+
- Future Motion Sequence
|
| 21 |
+
|
| 22 |
+
## Supported Actions
|
| 23 |
+
- walking
|
| 24 |
+
- running
|
| 25 |
+
- jumping
|
| 26 |
+
- sitting
|
| 27 |
+
- standing
|
| 28 |
+
- climbing
|
| 29 |
+
- crouching
|
| 30 |
+
- grabbing_object
|
| 31 |
+
- throwing
|
| 32 |
+
- pushing
|
| 33 |
+
- turning_left
|
| 34 |
+
- turning_right
|
| 35 |
+
|
| 36 |
+
## Hyperparameters
|
| 37 |
+
- Learning Rate: 0.0001
|
| 38 |
+
- Batch Size: 32
|
| 39 |
+
- Epochs: 100
|
| 40 |
+
- Optimizer: AdamW
|
| 41 |
+
- Dropout: 0.15
|
| 42 |
+
- Activation: GELU
|
| 43 |
+
- Loss Function: CrossEntropy + SmoothL1
|
| 44 |
+
|
| 45 |
+
## Dataset
|
| 46 |
+
- 25,000 motion sequences
|
| 47 |
+
- 12 action classes
|
| 48 |
+
- Augmented with noise, rotation, scaling, and time distortion
|
| 49 |
+
|
| 50 |
+
## License
|
| 51 |
+
Apache 2.0
|