Robotics · 2026-10-05
AVL-JEPA: Preventing Causal Dynamics Information Collapse In Joint Embedding Predictive Architecture World Models
Yikang Qiao, Ling Zhang, Ziying Song, Duan Huang
AI-assisted summary based on the paper. Consult the linked source for methods and limitations. Benchmark numbers appear after editor verification.
TL;DR
The authors report AVL, an action-grounded vision-invariance latent method for JEPA world models that targets “causal dynamics information collapse.” They use executed actions as an auxiliary anchor and align clean vs. perturbed latent transitions while freezing the action head. On four robotic control tasks (TwoRoom, PushT, OGBench Cube, Reacher), AVL improves success under visual perturbations while preserving clean performance.
Why it matters
JEPA-style world models can avoid global representation collapse yet still fail to preserve action-conditioned causal dynamics needed for planning. The paper’s AVL is designed to prevent this specific failure mode and shows improved planning robustness with mechanism-level evidence (clean–noisy consistency, consequence alignment, targeted erasure).
Method
- Propose AVL to prevent causal dynamics information collapse in JEPA world models using (1) an action-grounded pathway that anchors latent transition responses to executed actions, and (2) a vision-invariance pathway that aligns clean vs. perturbed latent predictions while freezing the action head.
- Action-grounded pathway builds a transition-attributed action head by subtracting a zero-transition (state-only) baseline, and uses an auxiliary action prediction loss alongside JEPA prediction and SIGReg.
- Vision-invariance pathway perturbs only the current observation (clean future target), keeps the action head frozen, and maintains the action loss active so alignment cannot be achieved by discarding dynamics information.
Limitation
AVL builds on the SIGReg regularizer of LeWM rather than replacing it, and the authors evaluate it only in simulation. The authors plan real robot experiments in future work.
Abstract (from arXiv)
Joint embedding predictive architectures (JEPAs) predict future latent representations without reconstructing observations, enabling world models to focus on high-level semantic dynamics. However, a JEPA can preserve high dimensional visual information while discarding information about the physical consequences of actions. We call this failure mode causal dynamics information collapse and propose action-grounded vision-invariance latent (AVL) to prevent this collapse. We first use the executed action as an auxiliary dynamics anchor that encourages the model to preserve dynamics information, and then use a vision-invariance pathway which aligns perturbed and clean latent predictions without discarding dynamics information, forcing the model to fully understand and utilize causal dynamics information. We validate AVL on four robotic control tasks (TwoRoom, PushT, OGBench Cube, and Reacher), showing that it substantially improves success rates under visual perturbations while preserving clean-environment performance. We further evaluate physical consequence alignment, clean-noisy dynamics consistency, and the causal effect of targeted transition subspace erasure. Collectively, these results indicate that dynamic information causally relevant to planning is preserved from collapse under AVL.
Related papers
- Preserving Unstable Modes Through Inverse Dynamics in JEPA World Models
- Juno: Taming Predictive Latents for Vision-Language-Action Models
- Hamiltonian JEPA: Action-Conditioned World Models with an Inherited Control State
- SplitJEPA: Learning Invariant and Variant Latent Worlds without Reconstruction
- Causal Analysis of Robot Foundation Models Enables Efficiency Gains
Publication record
Bibliographic evidence describes where and when a record appeared. It does not validate scientific claims.
First public release: 2026-10-02 · Indexed by Pub-AI: 2026-10-05 UTC
AVL-JEPA: Preventing Causal Dynamics Information Collapse In Joint Embedding Predictive Architecture World Models
Preprint posted: 2026-10-02 · Original record ↗
- arxiv
- 2610.03587
Explicit source versions (1)
- v1 · 2026-10-02
- Preprint available
Source date: 2026-10-02 · First detected: 2026-10-25 UTC · Historical baseline
Source evidence (1)
- Official evidence ↗ · atom:published (official metadata)
Abstract · arxiv
Joint embedding predictive architectures (JEPAs) predict future latent representations without reconstructing observations, enabling world models to focus on high-level semantic dynamics. However, a JEPA can preserve high dimensional visual information while discarding information about the physical consequences of actions. We call this failure mode causal dynamics information collapse and propose action-grounded vision-invariance latent (AVL) to prevent this collapse. We first use the executed action as an auxiliary dynamics anchor that encourages the model to preserve dynamics information, and then use a vision-invariance pathway which aligns perturbed and clean latent predictions without discarding dynamics information, forcing the model to fully understand and utilize causal dynamics information. We validate AVL on four robotic control tasks (TwoRoom, PushT, OGBench Cube, and Reacher), showing that it substantially improves success rates under visual perturbations while preserving clean-environment performance. We further evaluate physical consequence alignment, clean-noisy dynamics consistency, and the causal effect of targeted transition subspace erasure. Collectively, these results indicate that dynamic information causally relevant to planning is preserved from collapse under AVL.