inr2action: Encoding Robot Episodes as Neural Implicit Representations for Recall and Action

Key points

• Long-horizon robotic operation requires memory that scales with essential information rather than execution time.1
• A neural implicit representation is fitted to episode video from a shared, meta-learned initialization; the resulting weight delta constitutes the memory.1
• A learned projector maps the weight delta into token embeddings, conditioning a language model for question answering and a decoder for joint command regression, without frame retrieval at query time.1
• By representing individual experiences as fixed-size weight deltas in a common parameter space, this approach decouples memory cost from time horizon and provides a basis for scalable streaming memory.1

Sources (1)

• [1]inr2action: Encoding Robot Episodes as Neural Implicit Representations for Recall and Action
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 20, 08:00 PM

  We present a parameter-space memory framework where each episode is stored as a separable object.

  Long-horizon robotic operation requires memory that scales with essential information rather than execution time.

Extractive summary: sentences quoted from the sources.