aiqbench analysis
CLAMP addresses a critical gap in embodied AI by enforcing hard constraints during the decoding phase of vision-language models to ensure generated plans are physically executable and safe. By integrating scene evidence with symbolic action models, this framework prevents hallucinated object references and invalid state transitions without retraining the base model. The approach demonstrates that grounding language priors in real-time environmental observations significantly improves reliability for autonomous agents operating in complex, unseen environments.
CLAMP: Constrained Decoding for Vision-Language Embodied Planning
arXiv:2609.08602v1 Announce Type: new Abstract: Embodied planning increasingly relies on vision-language models (VLMs) to translate instructions and visual observations into executable action sequences. However, fluent plans are not always executable. A VLM may refer to objects that are not visually observed, select actions whose required affordances are unavailable, or violate syntax and action constraints. We introduce CLAMP, a multimodal constraint-grounding framework that turns scene evidence into decoding-time constraints for a frozen VLM planner. CLAMP uses the initial observation to restrict object references to those supported by the scene, while a provided symbolic action model specifies state transitions and goals. During decoding, hard masks eliminate invalid next-token candidates, while a Hidden Markov Model (HMM)-based world-state lookahead module reweights the probabilities of the remaining feasible candidates based on action preconditions and goal reachability. This allows the planner to retain the VLM's language prior while preventing visually unsupported, unsafe, or infeasible candidates from entering the plan. For unseen tasks and environments, CLAMP adapts the HMM at test time using label-free continuations sampled from the frozen VLM. Experiments on VLABench, SafeAgentBench, and TaPA show that scene-grounded constraints improve object grounding and safety, while most remaining failures stem from perception errors or misaligned constraint specifications.
Abstract:Embodied planning increasingly relies on vision-language models (VLMs) to translate instructions and visual observations into executable action sequences. However, fluent plans are not always executable. A VLM may refer to objects that are not visually observed, select actions whose required affordances are unavailable, or violate syntax and action constraints. We introduce CLAMP, a multimodal constraint-grounding framework that turns scene evidence into decoding-time constraints for a frozen VLM planner. CLAMP uses the initial observation to restrict object references to those supported by the scene, while a provided symbolic action model specifies state transitions and goals. During decoding, hard masks eliminate invalid next-token candidates, while a Hidden Markov Model (HMM)-based world-state lookahead module reweights the probabilities of the remaining feasible candidates based on action preconditions and goal reachability. This allows the planner to retain the VLM's language prior while preventing visually unsupported, unsafe, or infeasible candidates from entering the plan. For unseen tasks and environments, CLAMP adapts the HMM at test time using label-free continuations sampled from the frozen VLM. Experiments on VLABench, SafeAgentBench, and TaPA show that scene-grounded constraints improve object grounding and safety, while most remaining failures stem from perception errors or misaligned constraint specifications.
Artificial Intelligence (cs.AI)
arXiv:2609.08602 [cs.AI]
(or arXiv:2609.08602v1 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2609.08602 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 289 words
Related coverage
Neurosymbolic Embodied Agents
arXiv · Aug 18, 2026
Build a Physical AI model factory with NVIDIA Cosmos 3 on SageMaker HyperPod
AWS · Sep 4, 2026
[AINews] not much happened today
Latent Space · Sep 18, 2026
Fast Plans, Faithful Actions: Closing the Planning-Execution Gap in Hierarchical Vision-Language-Action Models
arXiv · Sep 28, 2026