Omega Humanoid AI v4

Overview

Model Details

• Architecture: TCN + Transformer Encoder
• Framework: PyTorch
• Tasks: Action Classification, Motion Forecasting, Balance Optimization
• Number of Classes: 12

Input Format

• Sequence Length: 120 frames
• Joints: 28
• Features per Joint: 6 (x, y, z, vx, vy, vz)

Output

• Action Label
• Confidence Score
• Future Motion Sequence

Supported Actions

• walking
• running
• jumping
• sitting
• standing
• climbing
• crouching
• grabbing_object
• throwing
• pushing
• turning_left
• turning_right

Hyperparameters

• Learning Rate: 0.0001
• Batch Size: 32
• Epochs: 100
• Optimizer: AdamW
• Dropout: 0.15
• Activation: GELU
• Loss Function: CrossEntropy + SmoothL1

Dataset

• 25,000 motion sequences
• 12 action classes
• Augmented with noise, rotation, scaling, and time distortion

License