ATM pre-trains models to predict trajectories of any points in videos, then uses those predictions to learn strong visuomotor policies from minimal action labels, beating baselines by 80% on 130+ tasks.
Diffusion Policy: Visuomotor Policy Learning via Action Diffusion
10 Pith papers cite this work, alongside 411 external citations. Polarity classification is still indexing.
10
Pith papers citing it
411
external citations · OpenAlex
citation-role summary
background 1
method 1
citation-polarity summary
representative citing papers
ReGuide is a self-improving framework that uses phase-conditioned guidance to generate corrective rollouts and absorbs successful ones back into diffusion policy training, yielding 1.3-7.7x success gains on Robomimic tasks.
BEACON uses discrepancy-aware importance reweighting to jointly train diffusion-based robot policies and source sample weights, improving performance over target-only and fixed-ratio baselines in cross-domain manipulation tasks.
EnergyFlow shows that denoising score matching on diffusion policies recovers the gradient of the expert's soft Q-function under maximum-entropy optimality, enabling non-adversarial reward extraction and improved policy generalization.
Projecting robot end-effector state onto image feature maps and sampling co-located visual tokens improves manipulation policy success by 4-10% across 67 tasks.
RoboLineage introduces an agent-native data lifecycle governance system that represents robot policy iteration steps as typed lineage artifacts to improve speed and auditability in real-robot workflows.
LUCID learns embodiment-agnostic intent models from unstructured human videos to train dexterous robot policies in simulation, enabling zero-shot transfer on real-world tasks like stirring and wiping.
DriveAnchor improves collision avoidance in autonomous driving planning via a three-stage anchor-based flow pipeline with pretraining on trajectory vocabulary, energy field post-training, and reward-refined fine-tuning.
ForceFlow improves success rates by 37% on six real-world contact-rich tasks over ForceVLA by treating force as a global regulatory signal in a flow-matching policy with hierarchical vision-to-force decomposition.
VLA-GSE uses spectral decomposition of the VLA backbone to create generalized and specialized experts, enabling effective robot task adaptation while updating only 2.51% of parameters and achieving 81.2% zero-shot success on LIBERO-Plus.
citing papers explorer
-
Any-point Trajectory Modeling for Policy Learning
ATM pre-trains models to predict trajectories of any points in videos, then uses those predictions to learn strong visuomotor policies from minimal action labels, beating baselines by 80% on 130+ tasks.
-
ReGuide: From Test-Time Guidance to Self-Improving Diffusion Policies
ReGuide is a self-improving framework that uses phase-conditioned guidance to generate corrective rollouts and absorbs successful ones back into diffusion policy training, yielding 1.3-7.7x success gains on Robomimic tasks.
-
BEACON: Cross-Domain Co-Training of Generative Robot Policies via Best-Effort Adaptation
BEACON uses discrepancy-aware importance reweighting to jointly train diffusion-based robot policies and source sample weights, improving performance over target-only and fixed-ratio baselines in cross-domain manipulation tasks.
-
Recovering Hidden Reward in Diffusion-Based Policies
EnergyFlow shows that denoising score matching on diffusion policies recovers the gradient of the expert's soft Q-function under maximum-entropy optimality, enabling non-adversarial reward extraction and improved policy generalization.
-
GeoProp: Grounding Robot State in Vision for Generalist Manipulation
Projecting robot end-effector state onto image feature maps and sampling co-located visual tokens improves manipulation policy success by 4-10% across 67 tasks.
-
RoboLineage: Agent-Native Data Lifecycle Governance Across Robot Policy Iterations
RoboLineage introduces an agent-native data lifecycle governance system that represents robot policy iteration steps as typed lineage artifacts to improve speed and auditability in real-robot workflows.
-
LUCID: Learning Embodiment-Agnostic Intent Models from Unstructured Human Videos for Scalable Dexterous Robot Skill Acquisition
LUCID learns embodiment-agnostic intent models from unstructured human videos to train dexterous robot policies in simulation, enabling zero-shot transfer on real-world tasks like stirring and wiping.
-
DriveAnchor: Progressive Anchor-based Flow Learning for Autonomous Driving Planning
DriveAnchor improves collision avoidance in autonomous driving planning via a three-stage anchor-based flow pipeline with pretraining on trajectory vocabulary, energy field post-training, and reward-refined fine-tuning.
-
ForceFlow: Learning to Feel and Act via Contact-Driven Flow Matching
ForceFlow improves success rates by 37% on six real-world contact-rich tasks over ForceVLA by treating force as a global regulatory signal in a flow-matching policy with hierarchical vision-to-force decomposition.
-
VLA-GSE: Boosting Parameter-Efficient Fine-Tuning in VLA with Generalized and Specialized Experts
VLA-GSE uses spectral decomposition of the VLA backbone to create generalized and specialized experts, enabling effective robot task adaptation while updating only 2.51% of parameters and achieving 81.2% zero-shot success on LIBERO-Plus.