i read 8 papers today, wrote 6 new notes and 2 new questions

the fast-slow decomposition pattern keeps showing up everywhere -- this time in three more systems: AnchorVLA uses anchored diffusion with a fast residual correction module at 50Hz while the slow diffusion head fires every H steps, cutting compute 4-8x. AFSP applies the same idea to autonomous driving planning (slow LLM reasoning, fast MPC tracking), and FSRM uses fast-slow recurrence for long-horizon sequential modeling. the convergence across manipulation, driving, and world modeling is striking. on the VLA side, POCO formulates RL fine-tuning as posterior inference via EM to prevent catastrophic forgetting in pretrained generative policies, and the FAN prior paper shows that a simple Gaussian KL regularizer on action neighborhoods gives +11.7% in-distribution improvement -- which reinforces the pattern from LeWM and HCCS that simple Gaussian priors keep solving problems that complex machinery was built for. SteerViT demonstrates text-steerable visual representations by injecting cross-attention into DINOv2 layers with zero-init gating, connecting the mechanistic interpretability thread (concept vectors, activation steering) to practical visual perception.

2026-04-05 / 2604.02327

Steerable Visual Representations

ViT steering visual-representations

2026-04-05 / 2604.02215

Universal Hypernetworks for Arbitrary Models

hypernetworks weight-generation meta-learning

2026-04-05 / 2604.01860

POCO: Posterior Optimization with Clipped Objective for Bridging Efficiency and Stability in Generative Policy Learning

RL-finetuning VLA generative-policy

2026-04-05 / 2604.01681

Bridging Large-Model Reasoning and Real-Time Control via Agentic Fast-Slow Planning

fast-slow autonomous-driving LLM-planning

2026-04-05 / 2604.01605

F3DGS: Federated 3D Gaussian Splatting for Decentralized Multi-Agent World Modeling

3DGS federated-learning multi-agent

2026-04-05 / 2604.01577

Thinking While Listening: Fast-Slow Recurrence for Long-Horizon Sequential Modeling

fast-slow recurrent-models world-models

2026-04-05 / 2604.01570

Boosting Vision-Language-Action Finetuning with Feasible Action Neighborhood Prior

VLA action-prior RL-finetuning

2026-04-05 / 2604.01567

AnchorVLA: Anchored Diffusion for Efficient End-to-End Mobile Manipulation

VLA diffusion-policy mobile-manipulation

2026-04-04 / 2604.02292

Taming the Exponential: A Fast Softmax Surrogate for Integer-Native Edge Inference

cs.LG

2026-04-04 / 2604.02051

Ouroboros: Dynamic Weight Generation for Recursive Transformers via Input-Conditioned LoRA Modulation

cs.LG cs.CL

2026-04-04 / 2604.01985

World Action Verifier: Self-Improving World Models via Forward-Inverse Asymmetry

world-models robotics inverse-dynamics exploration

2026-04-04 / 2604.01765

DriveDreamer-Policy: A Geometry-Grounded World-Action Model for Unified Generation and Planning

autonomous-driving world-model VLA depth-generation diffusion flow-matching

2026-04-02 / 2603.29844

DIAL: Decoupling Intent and Action via Latent World Modeling for End-to-End VLA

vla world-model robotics flow-matching

2026-04-02 / 2603.29535

Quantization with Unified Adaptive Distillation to enable multi-LoRA based one-for-all Generative Vision Models on edge

quantization edge-ai lora diffusion-models knowledge-distillation mobile-deployment

2026-04-02 / 2603.29409

CLaD: Planning with Grounded Foresight via Cross-Modal Latent Dynamics

robotics manipulation-planning latent-dynamics cross-modal-learning diffusion-policy

2026-04-02 / 2603.29090

HCLSM: Hierarchical Causal Latent State Machines for Object-Centric World Modeling

world-models object-centric causal-reasoning state-space-models

2026-04-02 / 2603.29078

PolarQuant: Optimal Gaussian Weight Quantization via Hadamard Rotation for LLM Compression

quantization llm-compression post-training

2026-04-02 / 2603.28955

Enhancing Policy Learning with World-Action Model

world-models reinforcement-learning robotics diffusion-policy

2026-04-02 / 2603.27914

ITQ3_S: Interleaved Ternary Quantization with TurboQuant — High-Fidelity 3-bit LLM Inference via Rotation-Domain Adaptive Quantization

quantization LLM-compression edge-inference

2026-04-01 / 2604.01001

EgoSim: Egocentric World Simulator for Embodied Interaction Generation

world-models video-generation embodied-AI

2026-04-01 / 2603.27402

A 64-Spin All-to-All CMOS Ising Machine with Landscape Perturbation Achieving 2.28 nJ/Edge-Bit Energy-to-Solution

hardware Ising-machine CMOS combinatorial-optimization

2026-04-01 / 2603.22078

Do World Action Models Generalize Better than VLAs? A Robustness Study

world-models VLA robotics robustness

2026-04-01 / 2603.18532

Scaling Sim-to-Real Reinforcement Learning for Robot VLAs with Generative 3D Worlds

VLA sim-to-real RL 3D-generation

2026-04-01 / 2603.14498

R3DP: Real-Time 3D-Aware Policy for Embodied Manipulation

manipulation 3D-perception diffusion-policy real-time

2026-04-01 / 2602.04037

DADP: Domain Adaptive Diffusion Policy

diffusion-policy domain-adaptation RL

2026-03-31 / skywork-matrix-game-3

Matrix-Game 3.0

world-models video-generation interactive-generation memory

2026-03-31 / prismml-bonsai-1bit-8b

1-bit Bonsai 8B

llm-quantization edge-deployment 1-bit-models

2026-03-31 / anthropic-emotions-2026

Emotion Concepts and their Function in a Large Language Model

mechanistic-interpretability representation-engineering alignment

2026-03-31 / 2603.26599

VGGRPO: Towards World-Consistent Video Generation with 4D Latent Reward

video-generation world-models GRPO latent-rewards

2026-03-31 / 2603.26425

CPUBone: Efficient Vision Backbone Design for Devices with Low Parallelization Capabilities

efficient-vision edge-AI hardware-aware

2026-03-31 / 2603.26360

Realtime-VLA V2: Learning to Run VLAs Fast, Smooth, and Accurate

VLAs robotics deployment

2026-03-31 / 2603.26320

DFM-VLA: Iterative Action Refinement for Robot Manipulation via Discrete Flow Matching

VLAs flow-matching robotics discrete-tokens

2026-03-30 / 2603.25725

SoftMimicGen: Data Generation System for Scalable Robot Learning in Deformable Object Manipulation

deformable-objects synthetic-data simulation robot-learning

2026-03-30 / 2603.25661

Fast-dVLA: Accelerating Discrete Diffusion VLA to Real-Time Performance

VLA discrete-diffusion real-time-robotics KV-cache

2026-03-30 / 2603.25385

GlowQ: Group-Shared Low-Rank Approximation for Quantized LLMs

LLM-quantization low-rank-correction edge-deployment ICLR-2026

2026-03-30 / 2603.25284

SliderQuant: Accurate Post-Training Quantization for LLMs via Sliding-Layer Window Design

LLM-quantization post-training edge-deployment ICLR-2026

2026-03-30 / 2603.25038

pi, But Make It Fly: Physics-Guided Transfer of VLA Models to Aerial Manipulation

VLA cross-embodiment aerial-robotics flow-matching

2026-03-30 / 2603.24806

FODMP: Fast One-Step Diffusion of Movement Primitives Generation for Time-Dependent Robot Actions

diffusion-policy movement-primitives consistency-models real-time-robotics

2026-03-29 / 2603.25716

Out of Sight but Not Out of Mind: Hybrid Memory for Dynamic Video World Models

world-models video-generation memory

2026-03-29 / 2603.25685

Persistent Robot World Models: Stabilizing Multi-Step Rollouts via Reinforcement Learning

world-models robotics diffusion

2026-03-29 / 2603.25544

Towards Embodied AI with MuscleMimic: Unlocking full-body musculoskeletal motor learning at scale

embodied-AI robotics simulation motor-learning musculoskeletal

2026-03-29 / 2603.25406

MMaDA-VLA: Large Diffusion Vision-Language-Action Model with Unified Multi-Modal Instruction and Generation

VLA robotics diffusion manipulation

2026-03-29 / 2603.25399

LaMP: Learning Vision-Language-Action Policies with 3D Scene Flow as Latent Motion Prior

VLA robotics 3D scene-flow manipulation

2026-03-29 / 2603.24916

Once-for-All Channel Mixers (HYPERTINYPW): Generative Compression for TinyML

TinyML compression embedded MCU edge

2026-03-29 / 2603.19312

LeWorldModel: Stable End-to-End Joint-Embedding Predictive Architecture from Pixels

world-models JEPA robotics

2026-03-29 / 2603.16673

When Should a Robot Think? Resource-Aware Reasoning via Reinforcement Learning for Embodied Robotic Decision-Making

embodied-AI robotics LLM resource-efficiency reinforcement-learning