feat: Add Transformer-based haptic encoder with multi-task learning

[Rakshitha-Ireddi]

Authors

• Ireddi Rakshitha
• Yashwanth Devavarapu

Summary

• Add TransformerHapticEncoder with temporal attention and MC Dropout uncertainty
• Add CrossModalAttention and HierarchicalFusion for improved modality fusion
• Add MultiTaskVisuoHapticModel for joint material/hardness/temperature prediction
• Add comprehensive haptic time series augmentation utilities
• Add training scripts with uncertainty-weighted multi-task loss

This contribution addresses the need for:

1. Attention-based temporal modeling (vs. InceptionTime convolutions)
2. Interpretable cross-modal fusion (vs. simple concatenation)
3. Multi-task learning for richer scene understanding
4. Uncertainty quantification for safe robotic decision-making