mlsanityData to 14 Oct 2026

Minimal Recurrent Behavioral Memory for Imitation under Partial Observability

arXiv 2609.25757 ↗First posted 22 Sep 2026cs.LG cs.IT cs.ROcited by 0

Not on the map. Why?

Abstract (arXiv)

What is the least recurrent memory needed to reproduce a specified expert under partial observability? The instantaneous requirement is the conditional entropy of the expert's behavioral quotient, but recurrence must also preserve distinctions that future observations will not restore before use. We characterize this minimal recurrent behavioral memory by a compatibility relation: under transitivity its classes attain the exact minimum, while the general case is an entropy minimization over closed compatible state assignments, with exact certificates on finite instances. A sole-carrier measurement protocol separates behavioral sufficiency, excess code rate, and information carried by observations or other memory paths; experimental bit requirements refer to the induced symbolic behavioral model under the stated occupancy. Across manipulation tasks, learned code rates remain near zero- and two-bit requirements as hidden modes grow to 512, and anticipatory memory follows a 2→1→0 requirement despite zero instantaneous demand during waiting. Learning this representation remains difficult: event-agnostic future-behavior supervision yields 36/40 sufficient seeds with one frozen configuration and improves the longest-horizon pixel setting from 0/8 to 6/8 sufficient held-out seeds (closed-loop success from 0.08 to 0.57). On unmodified community benchmarks, the protocol certifies delay-independent requirements, which sufficient codes match at mid-delay. The supervision aids commitment but can induce predictive surplus; annealing it lets imitation and rate training reduce that surplus, separating the information-theoretic target from the ability to learn it.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Builds on 37 of its 52 references that are in the map. First posted on arXiv 22 Sep 2026.

Builds on

37 of 52 references.

References · 37

  1. 2021RMA: Rapid Motor Adaptation for Legged Robotscited by 508
  2. 2020Learning Invariant Representations for Reinforcement Learning without Reconstructioncited by 436
  3. 2019Causal Confusion in Imitation Learningcited by 288
  4. 2024Behavior Generation with Latent Actionscited by 160
  5. 2017Soft-to-Hard Vector Quantization for End-to-End Learning Compressible Representationscited by 152
  6. 2019Scalable methods for computing state similarity in deterministic Markov Decision Processescited by 133
  7. 2018Variational Discriminator Bottleneck: Improving Imitation Learning, Inverse RL, and GANs by Constraining Information Flowcited by 105
  8. 2020Fighting Copycat Agents in Behavioral Cloning from Observation Historiescited by 71
  9. 2020Approximate information state for approximate planning and reinforcement learning in partially observed systemscited by 70
  10. 2023When Do Transformers Shine in RL? Decoupling Memory from Credit Assignmentcited by 63
  11. 2023POPGym: Benchmarking Partially Observable Reinforcement Learningcited by 62
  12. 2024Bridging State and History Representations: Understanding Self-Predictive RLcited by 54
25 more
  1. 2022Guaranteed Discovery of Control-Endogenous Latent States with Multi-Step Inverse Modelscited by 45
  2. 2025Memory, Benchmark & Robots: A Benchmark for Solving Complex Tasks with Reinforcement Learningcited by 36
  3. 2021Efficient Transformers in Reinforcement Learning using Actor-Learner Distillationcited by 35
  4. 2022Sequence Model Imitation Learning with Unobserved Contextscited by 34
  5. 2025Learning Long-Context Diffusion Policies via Past-Token Predictioncited by 30
  6. 2019Learning Belief Representations for Imitation Learning in POMDPscited by 24
  7. 2023Representation Learning with Multi-Step Inverse Kinematics: An Efficient and Optimal Approach to Rich-Observation RLcited by 22
  8. 2023Memory Gym: Towards Endless Tasks to Benchmark Memory Capabilities of Agentscited by 16
  9. 2023Rapid Motor Adaptation for Robotic Manipulator Armscited by 14
  10. 2026Gated Memory Policy: In-Context Memorization and Adaptationcited by 14
  11. 2026Memory Retrieval in Visuomotor Policies for Long-Horizon Robot Controlcited by 8
  12. 2025Benchmarking Partial Observability in Reinforcement Learning with a Suite of Memory-Improvable Domainscited by 7
  13. 2026Remember the Decision, Not the Description: A Rate-Distortion Framework for Agent Memorycited by 7
  14. 2024Learning Memory Mechanisms for Decision Making through Demonstrationscited by 6
  15. 2026The Myhill-Nerode Theorem for Bounded Interaction: Canonical Abstractions via Agent-Bounded Indistinguishabilitycited by 3
  16. 2016Minimum-Information LQG Control - Part II: Retentive Controllerscited by 2
  17. 2024Generalizing Multi-Step Inverse Models for Representation Learning to Finite-Memory POMDPscited by 2
  18. 2026Training and Evaluating Diffusion Policies with Long Context Lengthscited by 2
  19. 2026Minimal Markovization via Stable Quotients in Holonomy-Cover Decision Processescited by 2
  20. 2026On the Capability Separation Between World-Model Policy Learning and Imitated World-Action Modelscited by 2
  21. 2016Memory Lens: How Much Memory Does an Agent Use?cited by 1
  22. 2026Support sufficiency as action-sufficient compression: a single-cycle rate-regret formulationcited by 1
  23. 2026What Must Generalist Agents Remember?cited by 1
  24. 2026Remember what you did?: Learning Behavioral Memories for Partially Observable Object Manipulationcited by 1
  25. 2026Compressing History into Memory: Distilling Transformers into Recurrent Transformerscited by 1

Built on by

Not cited yet.