Research Papers
New Survey Maps How Multimodal AI Agents Remember and Evolve
A Hugging Face survey proposes a two-axis framework for multimodal agent memory, separating how information is formed and represen...
CriticOPD Fixes On-Policy Distillation by Letting Teachers Critique Student Errors
Hugging Face researchers introduce CriticOPD, a method that restores the teacher's ability to recover from student mistakes during...
ActionOPD: Distilling Agent Decisions, Not Tokens
A new Hugging Face paper proposes ActionOPD, an action-level on-policy distillation method that aligns a student agent's decisions...
LatentLoop Rethinks Looped MoE Models With Asymmetric Compute Allocation
A new Hugging Face paper proposes LatentLoop, which compresses the first loop visit of a Mixture-of-Experts model, keeps the secon...
Player-1 Trains Game Agents on Synthetic Experience, Beating Baselines
Hugging Face paper introduces Player-1, a post-training framework that extends computer-use agents with synthetic environments and...
Hugging Face Paper Unveils ATAS, a World-State Framework for Consistent Agentic Video
Any Time, Any Space (ATAS) compiles a screenplay into shot requests with explicit location, camera, and event-time references, dra...
GreenWAM: A Single World Foundation Model That Both Predicts and Acts Across 20 Robot Bodies
Hugging Face paper introduces GreenWAM, a robotics world foundation model that unifies action-conditioned world prediction and pol...
Tev-Gym and Tev-27B: Dense Rewards Push Long-Horizon AI Agents Forward
Hugging Face paper introduces Tev-Gym, a 1,170-environment long-horizon agent training dataset with verifiable dense progress rewa...
Urbex Turns City-Scale 3D Grounding Into an Agentic Search Task
A new Hugging Face paper, Urbex, reframes city-scale 3D grounding as active spatial search by a vision-language agent instead of r...
Agent-Reflex Teaches LLM Agents to Reflect and Recover in Noisy Tool Environments
A new training framework from RUC-NLPIR mines reflection skills from real tool failures and internalizes them into language agents...
VideoJev Turns Video Understanding Into Fast, Reusable Decisions
Hugging Face researchers introduce VideoJev, a family of 4B and 35B-A3B System One models that replace autoregressive answer decod...
Hugging Face Paper Introduces Jeva: A Decision-Native "System One" Architecture
A new Hugging Face paper proposes Jeva, a decision-native architecture that jointly addresses null awareness, permutation invarian...