Latent Space Clustering of Visual Latent Representation
Investigating task interference, subgoal structure, and scene alignment in long-horizon visual representations.
This ongoing research studies how shared-scene multi-task training shapes latent-space clustering and task interference in visual representations.
Key directions:
- Analyzing latent phase transitions across subgoals in long-horizon robotic tasks.
- Understanding how scene structure changes representation consistency across tasks.
- Proposing scene alignment and task grouping strategies that improve generalization.