Neurosymbolic Embodied Agents

Source

• Source name: arXiv
• Source kind: paper
• Authors / organization: Mohammad Albinhassan, Yuming Feng, Alessandra Russo, Pranava Madhyastha
• Venue: arXiv: cs.RO
• Published: 2026-08-17
• Collected: 2026-08-18
• URL: http://arxiv.org/abs/2608.16794v1
• PDF: https://arxiv.org/pdf/2608.16794v1
• DOI: UNKNOWN
• Registry group: academic_apis_and_preprints
• Evidence role: canonical_academic_index
• Language: en
• Country / region: UNKNOWN

Plain-Language Brief

The authors propose a neurosymbolic agent that combines vision-language models with symbolic planning for executable long-horizon household tasks. The agent consists of two phases: the first involves a vision-language model and exploration harness to acquire goal-relevant information, and the second uses a PDDL transition model, Monte Carlo tree search, and a domain-independent planning heuristic to generate executable plans. The results show that this method outperforms direct visual policy in two environments (VirtualHome and ALFWorld) and solves more tasks when both constraints and search are used compared to using either alone.

What the Document Says

• The authors present a neurosymbolic agent for long-horizon household tasks.
• The agent consists of two phases: task-directed visual exploration and symbolic planning.
• In the first phase, a vision-language model and exploration harness acquire goal-relevant information.
• In the second phase, a PDDL transition model restricts decoding to applicable actions.
• Monte Carlo tree search evaluates executable continuations using a domain-independent planning heuristic.
• The resulting plans are executable by construction under the transition model.
• The method outperforms direct visual policy in two environments (VirtualHome and ALFWorld).
• In ALFWorld, either constraint or search alone solves under a third of tasks, but their combination solves over 95%.
• The method uses fewer generated tokens than extended thinking and fewer model-visible images than direct interaction.
• Residual failures are localized to state acquisition rather than plan generation.

Key Concepts Mentioned

• neurosymbolic agent
• long-horizon household tasks
• vision-language model
• PDDL transition model
• Monte Carlo tree search
• domain-independent planning heuristic
• VirtualHome
• ALFWorld

Claims Made by the Document

• The presented method outperforms direct visual policy in VirtualHome and ALFWorld.
• Combining constraints and search in ALFWorld solves over 95% of tasks compared to using either alone.

Evidence Used

• On VirtualHome and ALFWorld, open 4B-27B models exceed 90% success in both environments...
• Constraints and search prove complementary rather than interchangeable: in ALFWorld either alone solves under a third of tasks, whereas their combination solves over 95%.

Methods / Approach

UNKNOWN

Limitations Stated by the Authors

UNKNOWN

Limitations Noted by the Agent

Agent interpretation — not from the source.

UNKNOWN

Open Questions

UNKNOWN

Notes for Future Q&A

May include interpretation. Labeled where it does.

UNKNOWN

Raw Abstract / Excerpt

Language and vision-language models generate plausible embodied plans but do not guarantee executability, as their outputs can violate environment dynamics or act on incorrectly grounded entities. We present a neurosymbolic agent that factors long-horizon household tasks into task-directed visual exploration and constrained symbolic planning. In the first phase, a vision-language model and exploration harness acquire goal-relevant predicates and instance bindings from egocentric observations and grounded interactions, producing a symbolic initial state. In the second, a PDDL transition model restricts decoding to tokens that extend applicable actions. Monte Carlo tree search then evaluates executable continuations using a domain-independent planning heuristic. The resulting plans are executable by construction under the transition model, with transfer to the environment conditioned on correct visual grounding. On VirtualHome and ALFWorld, open 4B-27B models exceed 90% success in both environments, and our smallest agent substantially outperforms a 27B direct visual policy in each. Constraints and search prove complementary rather than interchangeable: in ALFWorld either alone solves under a third of tasks, whereas their combination solves over 95%. The method also uses several times fewer generated tokens than extended thinking and far fewer model-visible images than direct interaction, and residual failures localize to state acquisition rather than plan generation without any specialized training.