role decoupling
Imagine a system where one part handles deciding when to act and another handles what to do. Role decoupling means splitting different tasks, like planning versus executing, or controlling timing versus generating content, so each part can focus on its own job without interfering. This separation often leads to better overall performance, such as faster responses or deeper reasoning.
Across these studies, role decoupling consistently helps systems overcome trade-offs. It allows spoken agents to be both responsive and intelligent, video models to generate endlessly without extra training, and agentic systems to make more reliable decisions by separating reasoning from action. It also reduces costs and improves accuracy in tasks like long-context question answering and web navigation. The common finding is that dividing responsibilities enables each component to specialize, leading to better outcomes than keeping all functions in one module.
- Papers
- 48
- Released code
- 2
- First seen
- Aug 2026
- Latest
- Nov 2026
45 papers in the last two months, against 3 in the two before.
Start here
Three papers to read first, in the order they came out- 1Simple prompt trick stops AI agents from faking progress in robot tasks
Splitting robot reward estimation into a factual scene description followed by rubric-based scoring removes vision-language models' bias toward declaring failed tasks successful.
Nov 2026 - 2Motion model understands and generates in one shared space
Möbius unifies motion understanding and generation by diffusing noisy motion directly in the same multimodal space as text.
Nov 2026 - 3Robots maintain control speed as their AI models grow tenfold
Decoupling the vision-language backbone from the action head lets humanoid control rate stay at 20 Hz regardless of model size.
Nov 2026
How fast it is moving
Often studied with
credit assignmentDetermining which actions or steps in a sequence contributed to a final outcome, often for reinforcement learning.11 togetherreinforcement learningA machine learning method where an agent learns to make sequences of decisions by receiving rewards or penalties for its actions.7 togethermulti-agent architectureA system where multiple AI agents collaborate, each handling specialized roles, communicating and coordinating to complete complex tasks.7 togetherWho is working on it
The papers
Most central to this idea first, not most recent.
- Industrycs.AI
Robots can act at 50 Hz while updating visual understanding once a second
Leap-VLA decouples slow semantic from fast reactive control, reusing cached VLM context to match same-frequency performance.
Leapmotor · Nov 2026
- Top Universitycs.RO
Robots maintain control speed as their AI models grow tenfold
Decoupling the vision-language backbone from the action head lets humanoid control rate stay at 20 Hz regardless of model size.
XPENG Robotics, University of Washington · Nov 2026
- Big Techcs.RO
Simple prompt trick stops AI agents from faking progress in robot tasks
Splitting robot reward estimation into a factual scene description followed by rubric-based scoring removes vision-language models' bias toward declaring failed tasks successful.
University of Washington, NVIDIA · Nov 2026
- Chinese Techcs.CV
Motion model understands and generates in one shared space
Möbius unifies motion understanding and generation by diffusing noisy motion directly in the same multimodal space as text.
The Hong Kong University of Science and Technology, Tencent Games Motus · Nov 2026
- Big Techcs.CV
Learning visual concepts before rendering improves image generation in controlled tests
Visual generation is better when autoregressive pretraining learns semantic tokens and diffusion post-training renders them.
University of California, San Diego, Adobe Research · Oct 2026
- Big Techcs.AI
Separating coordination from execution improves prompt optimization in multi-agent AI systems
Separating shared coordination instructions from agent-specific execution prompts, with role-specific optimization guidance, improves multi-agent prompt optimizers across evaluated tasks.
Korea University, KAIST · Oct 2026
- Chinese Techcs.CV
Video generators follow user-defined rules better when separate software controls object behavior
BoxWorld separates interaction reasoning from generative rendering to create real-time interactive video worlds with user-defined behaviors.
University of British Columbia, Vector Institute for AI · Oct 2026
- Chinese Techcs.AI
Language models can compress conversation memory 6.5-fold with little accuracy loss
Quantizing the full MLA cache to about 2.5 bits per element with role-aware transforms stays nearly lossless on reasoning and coding across four MLA models.
University of Chinese Academy of Sciences, Institute of Automation, Chinese Academy of Sciences · Oct 2026
- Big Techcs.CL
Diffusion separates response timing from content in full-duplex speech
Diffuplex decouples when to speak from what to speak, enabling fast and intelligent full-duplex spoken dialogue.
NVIDIA · Oct 2026
- Big Techcs.AI
Separating agent checks from reasoning improves weaker models’ reliability
DeReAct improves AI agent reliability by moving action approval and task completion checks out of the main reasoning policy.
Amazon, Apple · Oct 2026
- Chinese Techcs.AI
Small test-time-trained guidance model beats adapting large LLM generators for discovery
Guidance-TTT separates strategic guidance from solution execution, using a small test-time trained model to outperform prior methods.
Massachusetts Institute of Technology, Hong Kong Polytechnic University · Oct 2026
- Academiccs.RO
Separating exploration and exploitation policies with classifier-free guidance improves diffusion RL fine-tuning
CFG-Policy combines separate exploration and exploitation diffusion policies via classifier-free guidance to improve RL performance while preserving behavioral diversity.
ShanghaiTech University, InstAdapt · Oct 2026
- Top Universitycs.CV
Separating context-writing and denoising parameters improves autoregressive video generation
Separating parameters for context writing and denoising in autoregressive video generation improves visual quality and long-horizon consistency without extra data or fine-tuning.
Tsinghua University, JD · Oct 2026
- Chinese Techcs.CL
Separating planning from synthesis improves long-horizon search agents
IterSynth separates planning and synthesis in deep search agents using a summary state, outperforming prior small agents.
Zhejiang University, Tencent · Sept 2026
- Big Techcs.CL
Distributed grounding preserves reasoning across million-token language model contexts
DISCO separates parallel, sharded evidence grounding from centralized reasoning, preserving long-context accuracy while cutting inference cost versus monolithic full-context models.
National University of Singapore, Adobe Research · Sept 2026
- Academiccs.AI
Decouple high-level VLM planning from low-level execution to speed up mobile agents
Separating VLM planning from lightweight, accessibility-based execution cuts mobile GUI agent time and API cost while retaining competitive task success.
Rice University · Sept 2026
- Academiccs.CR
Graph-based policy constrains LLM agents for topology-aware incident response
Sentinel-RL offloads topological reasoning from LLM agents to a PPO policy for reliable SOC incident response.
Indiana University · Sept 2026
- Academiccs.LG
Mid-depth layers judge KV-cache evictions better than the owning layer
Mid-depth layers score KV-cache eviction better than each layer scoring its own cache, improving long-context accuracy at small and mid-sized models.
Colby College · Nov 2026
- Big Techcs.AI
Chatbots can recover from earlier mistakes by rebuilding what users asked
Rebuilding the task specification from the user's messages alone, using an analyzer that never sees the assistant's turns, recovers most of the quality lost to polluted multi-turn context.
University of California, San Diego, Microsoft Research · Nov 2026
- Chinese Techcs.CV
Visual puzzle solvers improve by building explicit problem states
Explicit, revisable problem states, built from visual inspection and executed by generated programs, improve accuracy on visual reasoning benchmarks.
SenseTime Research, Tetras.AI · Nov 2026
- Chinese Techcs.SE
A two-agent system finds bugs better than one by splitting search and inspection
NINJA uses a navigator and inspectors to balance file-level and function-level search for better issue localization.
Zhejiang University, DAMO Academy, Alibaba Group · Nov 2026
- Big Techcs.AI
Online reinforcement learning on live iOS apps yields UI agents with 7x better cross-platform task success
Online RL with decoupled planning and grounding and dense LLM rewards lets small agents match frontier UI navigation models.
Apple · Oct 2026
- Big Techcs.LG
Detached shorter-horizon features boost long-horizon reinforcement learning
Separating the discount used for representation learning from the behavior discount improves deep RL performance across Atari and continuous control.
NVIDIA Research · Oct 2026
- Big Techcs.LG
Learning the search harness improves frozen models’ scientific problem solving
HarnessTTT learns to generate executable harnesses at test time, allowing a frozen small executor to outperform fine-tuned systems across scientific discovery tasks.
University of Wisconsin-Madison, NVIDIA · Oct 2026
- Research Labcs.RO
Hierarchical skill-based framework achieves high precision in robotic surgery by separating semantic decisions from low-level control
SurgiCLAW confines a multimodal language model to choosing atomic surgical skills while deterministic controllers execute them, achieving 94.6% accuracy and 88.0% real-robot success.
Institute of Automation, Chinese Academy of Sciences, University of Chinese Academy of Sciences · Oct 2026
- Industrycs.AI
Multi-agent LLM safety requires separate evaluations of communication, authorization, and execution
Multi-agent LLM safety depends separately on message delivery, signer authority, model requests, and tool enforcement, and valid authorization can still certify false claims.
Oct 2026
- Big Techcs.CV
Compact reusable memory improves question answering over hour-long videos
Video-EMem encodes long videos once into compact time-anchored text and sparse images, then retrieves evidence per query, improving accuracy and cutting latency.
Zhejiang University, Microsoft Research Asia · Oct 2026
- Chinese Techcs.AI
Interleaved teaching helps agents learn skills that transfer better
Interleaving expert teaching with student practice improves transferable skill learning for LLM agents.
Nanjing University, Alibaba Group · Oct 2026
- Big Techcs.CV
Separating visual and language reasoning phases improves vision-language model accuracy
SPLIT-RL separates visual and language reasoning training stages with claim-level advantages, improving accuracy over GRPO by 1.4-6.1 points.
University of Southern California, Amazon AGI · Oct 2026
- Chinese Techcs.CV
Parallel tile inspection helps models find details in large images
Visual Parallel Search with parallel tile inspection and adaptive zoom improves high-resolution VQA accuracy.
The University of Hong Kong, Huawei Research · Oct 2026
- Chinese Techcs.CR
Meta-attributes make agent tool permissions more consistent and attack-resistant
MetaPermit uses LLM-inferred meta-attributes and fixed policies to achieve zero malicious tool calls and up to 109% better task completion.
Paderborn University, Huawei Hilbert Research Center (Dresden) · Sept 2026
- Research Labcs.SE
Framework scales verifiable training environments for coding agents that build programs from scratch
ProgramGym's feedback-driven environment construction and role-isolated agent workflow improve from-scratch program reproduction, lifting Qwen3.8-27B to 55.71% on ProgramBench.
Xiamen University, Washington University in St. Louis · Oct 2026
- Top Universitycs.CRcode
Two-stage auditing finds more exploitable flaws in AI agent repositories
AgentXploit automates repository-to-runtime red-teaming for AI agents, achieving 59.3% end-to-end attack success.
National University of Singapore, University of North Carolina at Chapel Hill · Sept 2026
- Academiccs.CV
Splitting diffusion distillation into two phase-specific half-sized experts beats a full-size student.
Phase-wise Velocity Distillation uses two half-sized phase-specific experts to generate high-quality images in one full-backbone forward pass.
The Hong Kong Polytechnic University, OPPO Research Institute · Oct 2026
- Chinese Techcs.RO
Robots recover more often when vision-language models monitor failures
Spotter improves robot task success by letting embodied policies act continuously while a parallel vision-language model detects, diagnoses, and repairs failures.
Griffith University, IIIS, Tsinghua University · Sept 2026
- Chinese Techcs.CV
Splitting experts by role improves video diffusion model scaling
SplitMoE improves video diffusion by separating experts into semantic and generic groups to avoid spatiotemporal fragmentation.
University of Chinese Academy of Sciences, ByteDance · Oct 2026
- Independentcs.AI
External specifications, not agents, should authorize task completion claims
Specifications should govern agent execution and completion through an external runtime, not rely on agent self-assessment.
Sept 2026
- Chinese Techcs.CVcode
Token-aware latent refinement improves vision-language reasoning at inference time
Token-disentangled latent test-time scaling improves vision-language reasoning by routing visual and reasoning feedback to different token roles.
Qwen Business Unit of Alibaba, Nanjing University · Sept 2026
- Independentcs.AI
AI forecasting models take the top two spots across three benchmarks
Timestral's four time series foundation models set new state-of-the-art accuracy on multiple forecasting benchmarks.
Oct 2026
- Chinese Techcs.CL
Learning the refinement loop improves web generation beyond final pages
Learning generation, critique, and refinement jointly in a loop improves functional web generation more than optimizing each role in isolation.
Tsinghua University, Huawei Noah's Ark Lab · Oct 2026
- Big Techcs.AI
Agents search more effectively by choosing which reasoning branch to revisit
TreeFlex, where one model picks which tree node to resume and what to do there, beats flat ReAct and FoldAgent on research and SWE tasks.
Meta AI · Oct 2026
- Academiccs.LG
Single-pass backpropagation learns Pareto front at cost of standard training
A single forward-backward pass can recover a full front of Pareto stationary solutions in multi-objective deep learning.
Saarland University, Universidade Estadual de Campinas · Oct 2026
- Independentcs.AI
DSA orchestrates evidence-aware stock research across six regional markets
DSA is an evidence-aware orchestration framework for multi-market stock research with LLM agents.
Aug 2026
- Top Universitycs.AI
Lightweight decision model reduces token use in multi-agent systems by up to 54.6%
Replacing routing and review in multi-agent LLM workflows with a lightweight decision model cuts generative tokens by 12.7 to 54.6 percent while keeping accuracy within 1.34 points.
Shanghai Jiao Tong University · Oct 2026
- Big Techcs.LG
Shared memory improves looped Transformers while shrinking context caches
Shared memory across recursions improves quality and reduces memory in looped transformers.
IBM Research, Cornell University · Oct 2026
- Chinese Techcs.CL
Teacher-guided training improves specialization and coordination in multi-agent models
MAS-OPD achieves the highest mean score on code and math benchmarks by using on-policy distillation with role-advantage specialization and privileged attribution.
University of Science and Technology of China, Tencent · Sept 2026
- Chinese Techcs.CV
Executable code gives generative worlds persistent rules and evolving state
CoDeR combines explicit code-based world rules with diffusion video generation to enable open-ended world simulation.
USTC, TongYi Lab · Sept 2026
- Independentcs.AI
Deterministic guardrails expose gains that fool LLM-based judges
LLM judges are not reliable as final evaluators in self-improving loops; deterministic guardrails must override them.
Sept 2026
Concepts are extracted from each paper and reused across the corpus, so this page grows on its own as the desk reads.