newsBAIR (Berkeley)Trust 88 · LabPublished 3mo agoLive · 1mo ago
Gradient-based Planning for World Models at Longer Horizons
GRASP is a new gradient-based planner for learned dynamics (a “world model”) that makes long-horizon planning practical by (1) lifting the trajectory into virtual states so optimization is parallel
Why these links exist
Every edge carries a method, confidence, and the source snippet that justified it — so bad links are debuggable.
- LinkedLinked via unknownNot All Actions Are Equal: Rethinking Conditioning for Dexterous World Model →
- LinkedLinked via unknownPhysiFormer: Learning to Simulate Mechanics in World Space →
- LinkedLinked via unknownAutomating Potential-based Reward Shaping with Vision Language Model Guidance →
- LinkedLinked via unknownPhysisForcing: Physics Reinforced World Simulator for Robotic Manipulation →
- LinkedLinked via unknownAnyBody: Free-Form Whole-Body Humanoid Control from Arbitrary Keypoint Guidance →
- LinkedLinked via unknownZ-1: Efficient Reinforcement Learning for Vision-Language-Action Models →
- LinkedLinked via unknownAdaJEPA: An Adaptive Latent World Model →
- LinkedLinked via unknownValdi: Value Diffusion World Models →
Covers (incoming)
paperNot All Actions Are Equal: Rethinking Conditioning for Dexterous World ModelpaperPhysiFormer: Learning to Simulate Mechanics in World SpacepaperAutomating Potential-based Reward Shaping with Vision Language Model GuidancepaperPhysisForcing: Physics Reinforced World Simulator for Robotic ManipulationpaperAnyBody: Free-Form Whole-Body Humanoid Control from Arbitrary Keypoint GuidancepaperZ-1: Efficient Reinforcement Learning for Vision-Language-Action ModelspaperAdaJEPA: An Adaptive Latent World ModelpaperValdi: Value Diffusion World ModelspaperSequentially-Controlled Interactive Multi-Particle Flow-Maps for Online Feedback-Driven SearchpaperPhysMani: Physics-principled 3D World Model for Dynamic Object ManipulationpaperACID: Action Consistency via Inverse Dynamics for Planning with World ModelspaperWorldSample: Closed-loop Real-robot RL with World ModellingpaperSILO: Simulation-in-the-Loop Sim-to-Real Transfer for Multi-Stage Cable RoutingpaperDeform360: A Massive Multi-view Visuotactile Dataset for Deformable World ModelspaperGraph Sparse Sampling: Breaking the Curse of the Horizon in Continuous MDP PlanningpaperLearning to Throw Objects Safely in Multi-Obstacle EnvironmentspaperA Minimalist Retargeting-Guided Reinforcement Learning Recipe for Dexterous ManipulationpaperDirectional Constraints for Efficient Exploration in Safe Reinforcement LearningpaperUR-VC: Unsupervised Robotic Value Correction for Time-Derived Progress ProxiespaperSteering Robustness into World Action Models via Mechanistic Interpretability and Optimal ControlpaperConcept-Guided Spatial Regularization for World Models in Atari PongpaperPhysics-enhanced reinforcement learning for real-time optimal control of dynamical systemspaperS3: Stable Subgoal Selection by Constraining Uncertainty of Coarse Dynamics in Hierarchical Reinforcement Learning
Related across the graph
paperPhysMani: Physics-principled 3D World Model for Dynamic Object ManipulationpaperACID: Action Consistency via Inverse Dynamics for Planning with World ModelspaperS3: Stable Subgoal Selection by Constraining Uncertainty of Coarse Dynamics in Hierarchical Reinforcement LearningpaperPhysiFormer: Learning to Simulate Mechanics in World SpacepaperSequentially-Controlled Interactive Multi-Particle Flow-Maps for Online Feedback-Driven SearchpaperAutomating Potential-based Reward Shaping with Vision Language Model GuidancepaperUR-VC: Unsupervised Robotic Value Correction for Time-Derived Progress ProxiespaperSILO: Simulation-in-the-Loop Sim-to-Real Transfer for Multi-Stage Cable RoutingpaperPhysisForcing: Physics Reinforced World Simulator for Robotic ManipulationpaperAdaJEPA: An Adaptive Latent World ModelpaperSteering Robustness into World Action Models via Mechanistic Interpretability and Optimal ControlpaperLearning to Throw Objects Safely in Multi-Obstacle EnvironmentspaperAnyBody: Free-Form Whole-Body Humanoid Control from Arbitrary Keypoint GuidancepaperDeform360: A Massive Multi-view Visuotactile Dataset for Deformable World ModelspaperPhysics-enhanced reinforcement learning for real-time optimal control of dynamical systemspaperA Minimalist Retargeting-Guided Reinforcement Learning Recipe for Dexterous ManipulationpaperGraph Sparse Sampling: Breaking the Curse of the Horizon in Continuous MDP PlanningpaperZ-1: Efficient Reinforcement Learning for Vision-Language-Action ModelspaperWorldSample: Closed-loop Real-robot RL with World ModellingpaperNot All Actions Are Equal: Rethinking Conditioning for Dexterous World ModelpaperValdi: Value Diffusion World ModelspaperConcept-Guided Spatial Regularization for World Models in Atari PongpaperDirectional Constraints for Efficient Exploration in Safe Reinforcement Learning