Latent Dynamics · 2026-10-06
Mind the Execution Gap: Action-Semantic Mismatch in World-Model Control
Shengtao Wen, Xiang Chen, Yu Tian, Lingbing Guo, Lina Gong, Sheng-Jun 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 study how asynchronous action execution breaks the action semantics assumed by action-conditioned world-model controllers. They identify two architecture-dependent failure modes: TD-MPC2 suffers from a future-action timeline mismatch, while DreamerV3 can attribute transitions to previous actions that were not actually applied. They propose two execution-consistent interfaces—Future-Sequence and Applied-Action Feedback—that correct these mismatches without modifying pretrained world models.
Why it matters
World-model controllers use actions both for execution and as inputs to learned dynamics/RSSM updates. In real systems with delay, packet loss, reordering, and buffering, the commanded action may differ from the applied one, causing semantic misalignment inside the controller. The paper diagnoses where each architecture’s computation needs execution-consistent action information and provides lightweight interfaces to fix it without retraining the world models.
Method
- Controlled interventions isolate asynchronous execution from planner/model behavior by replaying planned commands from matched states while varying action-delivery delay, packet loss, and reordering.
- Diagnostic ablations test what execution information TD-MPC2 needs during latent rollout (current applied action/buffer vs exact future applied-action sequence) and what DreamerV3 needs for correct transition-action pairing (command vs applied-action feedback).
- The authors introduce execution-consistent interfaces: Future-Sequence for planning-based TD-MPC2 (transmit future action suffix and update an execution buffer) and Applied-Action Feedback for DreamerV3 (send an execution receipt / applied action aligned with each observed transition).
Abstract (from arXiv)
World-model controllers rely on action-conditioned dynamics for prediction and planning, yet real control systems often execute commands asynchronously due to communication delay, packet loss, reordering, and actuator buffering. We study how asynchronous execution changes the action semantics assumed within world-model controllers, rather than treating it only as an external control disturbance. Through controlled interventions, we identify two architecture-dependent failure modes: planning-based controllers such as TD-MPC2 suffer from a future-action timeline mismatch between imagined and executed action sequences, while recurrent world models such as DreamerV3 can attribute observed transitions to commands that were not actually applied. Our analysis shows that TD-MPC2 requires the correct future action sequence during latent dynamics rollout, whereas DreamerV3 requires timely attribution of each transition to the action that generated it. Based on these findings, we introduce two lightweight execution-consistent interfaces, Future-Sequence for TD-MPC2 and Applied-Action Feedback for DreamerV3, that correct these mismatches without modifying the pretrained world models. Experiments across delays, packet loss, reordering, multiple control domains, measured network traces, and a process-separated asynchronous stack consistently support both diagnoses and the corresponding architecture-specific corrections.
Related papers
- Plans Are Not Executions: Adapting Frozen World-Model Planners to Controller-Induced Execution Shift
- Plausible Future, Wrong Choice: Steering World-Action Models toward Intended Future
- Action-Consequence Alignment for Reliable Planning and Self-Improving in Latent World Models
- How Much Planning Is Enough? Reducing Search and Computation in World-Model Planning
- Commit While Futures Agree: Consequence-Aware Adaptive Action Chunking for Robot Manipulation
Publication record
Bibliographic evidence describes where and when a record appeared. It does not validate scientific claims.
First public release: 2026-10-05 · Indexed by Pub-AI: 2026-10-06 UTC
Mind the Execution Gap: Action-Semantic Mismatch in World-Model Control
Preprint posted: 2026-10-05 · Original record ↗
- arxiv
- 2610.06582
Explicit source versions (1)
- v1 · 2026-10-05
- Preprint available
Source date: 2026-10-05 · First detected: 2026-10-25 UTC · Historical baseline
Source evidence (1)
- Official evidence ↗ · atom:published (official metadata)
Abstract · arxiv
World-model controllers rely on action-conditioned dynamics for prediction and planning, yet real control systems often execute commands asynchronously due to communication delay, packet loss, reordering, and actuator buffering. We study how asynchronous execution changes the action semantics assumed within world-model controllers, rather than treating it only as an external control disturbance. Through controlled interventions, we identify two architecture-dependent failure modes: planning-based controllers such as TD-MPC2 suffer from a future-action timeline mismatch between imagined and executed action sequences, while recurrent world models such as DreamerV3 can attribute observed transitions to commands that were not actually applied. Our analysis shows that TD-MPC2 requires the correct future action sequence during latent dynamics rollout, whereas DreamerV3 requires timely attribution of each transition to the action that generated it. Based on these findings, we introduce two lightweight execution-consistent interfaces, Future-Sequence for TD-MPC2 and Applied-Action Feedback for DreamerV3, that correct these mismatches without modifying the pretrained world models. Experiments across delays, packet loss, reordering, multiple control domains, measured network traces, and a process-separated asynchronous stack consistently support both diagnoses and the corresponding architecture-specific corrections.