11 papers
Cortex: Consolidating Context into Parametric Memory
Jintian Zhang, Feng He, Xinnian Liang +5
Context can improve large language model performance, but these gains may be lost after context removal. Inspired by biological systems consolidation, we formulate Memory Consolida…
Schema: Discovering Unknown Environments via Agentic Program Induction
Guanning Zeng, Jiani Wang, Wenjie Ma +8
Learning to complete tasks in unfamiliar environments with unknown rules remains a key challenge for LLM agents. Current LLM agents often record their discoveries in prose, which m…
OmniTaskonomy: When Does Visual Generation Improve Visual Understanding?
Jiaxin Ge, Yiming Qin, Ji Xie +13
Training a model to generate visual content can encourage it to learn rich perceptual capabilities related to geometry, spatial relationships, and objectness; yet, its benefits for…
Tail-Likelihood Reinforcement Learning
Shrinivas Ramasubramanian, Daman Arora, Fahim Tajwar +11
Reinforcement learning typically optimizes average reward. For generative policies, the average can hide an important distinction: two policies can achieve the same mean reward whi…
T-Rex: Tactile-Reactive Dexterous Manipulation
Dantong Niu, Zhuoyang Liu, Zekai Wang +31
The ability to react dynamically to tactile signals has long been considered crucial to agile human-level dexterity. Yet contemporary learning-based Vision-Language-Action (VLA) mo…
Playful Agentic Robot Learning
Junyi Zhang, Jiaxin Ge, Hanjun Yoo +17
Current agentic robot systems can write executable Code-as-Policy programs, observe feedback, and revise behavior across multiple attempts, but they remain largely task-driven: reu…
DiPOD: Diffusion Policy Optimization without Drifting Apart
Haozhe Jiang, Haiwen Feng, Pieter Abbeel +3
RL post-training has become increasingly pivotal for improving diffusion policies, but existing diffusion policy-gradient methods are often unstable and cannot achieve reliable pol…
Maximum Likelihood Reinforcement Learning
Fahim Tajwar, Guanning Zeng, Yueer Zhou +7
Reinforcement learning (RL) is the method of choice for training models in setups where the objective function can only be evaluated by sampling from the model. Our key observation…
Visually Prompted Benchmarks Are Surprisingly Fragile
Haiwen Feng, Long Lian, Lisa Dunlap +6
A key challenge in evaluating VLMs is testing models' ability to analyze visual content independently from their textual priors. Recent benchmarks such as BLINK probe visual percep…
Vision-as-Inverse-Graphics Agent via Interleaved Multimodal Reasoning
Shaofeng Yin, Jiaxin Ge, Zora Zhiruo Wang +6
Vision-as-inverse-graphics, the concept of reconstructing images into editable programs, remains challenging for Vision-Language Models (VLMs), which inherently lack fine-grained s…
Flow Matching Policy Gradients
David McAllister, Songwei Ge, Brent Yi +5
Flow-based generative models, including diffusion models, excel at modeling continuous distributions in high-dimensional spaces. In this work, we introduce Flow Policy Optimization…