CondenseVLA: Learnable History Condensation for Efficient Multi-Frame VLA

Key points

• Robot manipulation often depends on short-term cues - such as brief contacts, temporary occlusions, and intermediate object states - that are ambiguous in a single frame and can only be resolved by integrating observations over time.1
• Given a sequence of cached per-frame features from the VLM backbone, CondenseVLA distills observation history into a fixed set of slot tokens using learnable queries and multi-head cross-attention.1
• Simple auxiliary regularization penalizes overlap among slot attention maps and decorrelates slot embeddings, encouraging complementary attention and reducing slot redundancy.1
• These results support selective condensation of bounded recent visual evidence for action decoding, while harder multi-observation tasks expose the limits of a finite history window.1

Sources (1)

• [1]CondenseVLA: Learnable History Condensation for Efficient Multi-Frame VLA
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 19, 07:26 AM

  We propose CondenseVLA, a multi-frame VLA policy with a learnable decoder-side history condensation module for using short-term visual context in action decoding.

  Robot manipulation often depends on short-term cues - such as brief contacts, temporary occlusions, and intermediate object states - that are ambiguous in a single frame and can only be resolved by integrating observations over time.

Extractive summary: sentences quoted from the sources.