OGBench
PulseAugur coverage of OGBench — every cluster mentioning OGBench across labs, papers, and developer communities, ranked by signal.
7 day(s) with sentiment data
-
New Horizon-Scalable Q-Learning method improves long-horizon off-policy value learning
Researchers have introduced Horizon-Scalable Q-Learning (HQL), a novel method designed to improve the accuracy of off-policy value learning in tasks with long horizons. HQL dynamically balances one-step and multi-step t…
-
New method BGP tackles path mirages in offline reinforcement learning
Researchers have introduced Branch-aware Graph Planning (BGP), a novel method designed to overcome challenges in offline goal-conditioned reinforcement learning, particularly in environments with high stochasticity and …
-
New CGSE method enhances offline learning in world models
Researchers have developed a novel method called Critic-Guided Synthetic Experience (CGSE) to enhance learning from offline datasets in world models. This approach uses the world model as a local transition generator, e…
-
New fine-tuning method preserves offline RL model performance without data retention
Researchers have developed a new method for fine-tuning reinforcement learning models that avoids retaining offline data. This approach, termed "Freeze, Don't Clone," addresses a critical failure mode where continued be…
-
New VRD Framework Tackles Out-of-Distribution Mode Averaging in Offline RL
Researchers have developed Value-Rectified Distillation (VRD), a new framework to address the challenge of Out-of-Distribution (OOD) mode averaging in offline Reinforcement Learning (RL). This problem occurs when distil…
-
New VPFlow method prioritizes value prediction for offline reinforcement learning
Researchers have developed VPFlow, a novel approach to offline goal-conditioned reinforcement learning that prioritizes value prediction over state prediction for planning. This method uses value regression as the sole …
-
VPFlow advances offline reinforcement learning with value-predictive planning
Researchers have developed VPFlow, a novel approach to offline goal-conditioned reinforcement learning that prioritizes value prediction over state prediction for planning. This method leverages a closed-form Monte Carl…
-
New TD-Aug method enhances robotic policy generalization via synthetic goal generation
Researchers have introduced TD-Aug, a novel method designed to enhance policy generalization in robotics by generating synthetic experience. This technique utilizes a geometric horizon model to imagine novel outcomes, e…
-
LASER algorithm enhances offline reinforcement learning with latent space control
Researchers have introduced LASER, a new offline reinforcement learning algorithm designed to improve policy optimization from static datasets. LASER addresses the challenge of out-of-distribution actions by constrainin…
-
New research explores higher-order action supervision and action shaping for robust RL policies
Two new arXiv papers explore advanced techniques in reinforcement learning (RL) and imitation learning (IL) to improve policy robustness and control. The first paper introduces a method for higher-order action supervisi…
-
New Q-learning method SQAM boosts performance on complex RL tasks
Researchers have developed a new reinforcement learning technique called Q-learning with Scalar Adjoint Matching (SQAM). This method addresses challenges in fine-tuning flow policies by introducing a closed-form scalar …
-
New ReQRL method enhances goal-conditioned reinforcement learning on OGBench
Researchers have developed ReQRL, a novel approach for goal-conditioned reinforcement learning (GCRL) that models goal-reaching costs as quasimetric distances. This method constrains the critic's value gradients using f…
-
Goal representation quality has minimal impact on GCRL performance, study finds
A new research paper investigates the impact of goal representations on goal-conditioned reinforcement learning (GCRL). The study found that the quality of the goal representation has minimal effect on downstream perfor…
-
Diffusion models advance with new architectures and training methods · 10 sources tracked
Researchers are exploring advanced diffusion models across various domains, including tri-modal generation for text, images, and audio by Apple, and data-free training methods for lattice field theories using diffusion …
-
ReBRAC-v2 advances offline reinforcement learning with simplified, high-performance methods
Researchers have introduced ReBRAC-v2, an updated approach to offline reinforcement learning that modernizes conventional actor-critic methods while maintaining algorithmic simplicity. This new method utilizes an exact-…
-
Research paper critiques contrastive critics in AI policy search
A new research paper titled "Good Rankers, Bad Objectives: Bilinear Contrastive Critics under Expressive Policy Search" explores the limitations of contrastive critics in AI policy search. The paper demonstrates that wh…
-
New HiQC Algorithm Enhances Offline RL for Long-Horizon Tasks
Researchers have introduced Hierarchical Implicit Q-Chunking (HiQC), a novel algorithm designed to improve offline goal-conditioned reinforcement learning for long-horizon tasks. HiQC addresses the 'curse of horizon' by…
-
New methods enhance diffusion policy training and inference speed
Researchers have developed new methods to improve the efficiency and stability of diffusion policies, a type of AI model gaining popularity for decision-making tasks. One approach, DIPOLE, introduces a novel reinforceme…
-
New DAGR method refines goal representations in reinforcement learning
Researchers have introduced DAGR, a novel approach to state-conditioned goal representations in reinforcement learning. DAGR refines existing goal embeddings by incorporating the current state through multi-scale gated …
-
MoP-JEPA introduces hard-assigned predictors for improved stochastic world models
Researchers have introduced MoP-JEPA, a novel approach to stochastic JEPA world models that addresses limitations in predicting successor states in environments with branching transitions. Unlike traditional JEPA models…