---
title: "Neurosymbolic Embodied Agents: Factoring Long-Horizon Household Tasks into Visual Exploration and Constrained Symbolic Planning"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2608.16794v1"
permalink: "https://feed.astralor.com/items/itm_0657259aa53a263d102242a1"
published: "2026-08-17T16:50:59.000Z"
score: 78
channel: "ai"
tags: ["PDDL","VirtualHome","VLM","Embodied AI","Neuro-symbolic","MCTS"]
---

# Neurosymbolic Embodied Agents: Factoring Long-Horizon Household Tasks into Visual Exploration and Constrained Symbolic Planning

## AI commentary

> What stands out is that this paper doesn't rely on bigger models; instead, it uses a neurosymbolic combination to make a 4B small model shine in embodied planning. For those working on robot decision-making or Agent deployment, this constrained decoding plus search approach may be a more efficient and controllable path than pure test-time compute.

## At a glance

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.

## Key points

- The neurosymbolic agent factors long-horizon household tasks into visual exploration and symbolic planning, first acquiring a symbolic initial state, then using PDDL-constrained decoding and MCTS search to guarantee executability.
- Open 4B-27B models exceed 90% success on both VirtualHome and ALFWorld, with the smallest agent substantially outperforming a 27B direct visual policy.
- Constraints and search are complementary: in ALFWorld, either alone solves under a third of tasks, while their combination solves over 95%.
- The method uses several times fewer generated tokens than extended thinking and far fewer model-visible images than direct interaction, with residual failures localized to state acquisition rather than plan generation.

## Full text

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.

---

Verify against the original before citing; curated by Astral Feed: https://feed.astralor.com/items/itm_0657259aa53a263d102242a1