World Models via Policy-Guided Trajectory Diffusion
...In this work, we propose a novel world modelling approach that is not autoregressive and generates entire on-policy trajectories in a single pass through a diffusion model....
https://arxiv.org/abs/2312.08533
World Models via Policy-Guided Trajectory Diffusion - OpenReview
...This paper introduces Policy-Guided Trajectory Diffusion, an approach that uses a world model to perform on-policy RL on trajectories ?in imagination?. Crucially, the proposed approach does not rely on autoregressive sampling but denoises full trajectories....
https://openreview.net/forum?id=9CcgO0LhKG
World Models via Policy-Guided Trajectory Diffusion (PolyGRAD)
...Official code to reproduce the experiments for the paper World Models via Policy-Guided Trajectory Diffusion. PolyGRAD diffuses an initially random trajectory of states and actions into an on-policy trajectory, and uses the synthetic data for imagined on-policy RL training....
https://github.com/marc-rigter/polygrad-world-models
[arxiv-cs.CV] ??-2025.11.26-??????? - ??
...????????????????-?-??? Uplifting Table Tennis: A Robust, Real-World Application for 3D Trajectory and Spin Estimation ?????????????????3D???????????????? Realizing Fully-Integrated, Low-Power, Event-Based Pupil Tracking with ......
https://zhuanlan.zhihu.com/p/1977050416605860041
[PDF] Policy-Guided Diffusion | Semantic Scholar
...Using synthetic experience from policy-guided diffusion as a drop-in substitute for real data, we demonstrate significant improvements in performance across a range of standard offline reinforcement learning algorithms and environments....
https://www.semanticscholar.org/paper/Policy-Guided-Diffusion-Jackson-Matthews/91fb21898301e80633d47d39b425d1feaa98f6ed
WorldModelsviaPolicy-GuidedTrajectoryDiffusion
...Abstract werful tool for developing intelligent agents. By predicting the outcome of a sequence of actions, world models enable policies to be optimised via on-policy reinforcement learning (RL) usin synthetic data, i.e. in ?in imagination?. Existing world models are autoregressive in that they interleave predicting the next state...
https://arxiv.org/pdf/2312.08533v2
WorldModelsviaPolicy-GuidedTrajectoryDiffusion
...Policy-Guided tRAjectory Difusion (PolyGRAD). A core novelty of PolyGRAD is that it enables the generation of entire on-policy trajectories in a single pass of difusion, rather than autoregressive...
https://openreview.net/pdf?id=9CcgO0LhKG
World models via policy-guided trajectory diffusion
...In this work, we propose a novel world modelling approach that is not autoregressive and generates entire on-policy trajectories in a single pass through a diffusion model....
https://ora.ox.ac.uk/objects/uuid:10be97c5-f4b1-412d-a98b-335fca4195a9
World Models via Policy-Guided Trajectory Diffusion - ????
...In this work, we propose a novel world modelling approach that is not autoregressive and generates entire on-policy trajectories in a single pass through a diffusion model....
https://www.zhuanzhi.ai/paper/fd864f005e12f417dcebd51a65fd60a4
World Models via Policy-Guided Trajectory Diffusion - arXiv.org
...In this work, we propose a novel world modelling approach that is not autoregressive and generates entire on-policy trajectories in a single pass through a diffusion model....
https://arxiv.org/html/2312.08533v4