APIVOT: Adaptive Planning with Interleaved Vision-Language Thoughts

2026-07-08

AI Summary

Long-horizon robotic planning requires balancing semantic task structure with geometric feasibility. To address this challenge, this work proposes APIVOT—a vision-language model (VLM)-based planner that introduces an adaptive interleaving mechanism between linguistic and visual reasoning. Specifically, language-based reasoning decomposes high-level tasks and sequences actions, while visual imagination generates and validates the geometric feasibility of anticipated future states. This joint optimization of semantic understanding and spatial constraint satisfaction enables APIVOT to significantly outperform existing VLM and planning frameworks in long-horizon kitchen tasks, particularly excelling in spatially constrained scenarios.

Abstract

Long-horizon robot planning requires jointly reasoning over semantic task structure and geometric feasibility. To successfully execute a task, a robot must decompose goals, select task-relevant objects, and sequence actions, while ensuring that plans satisfy spatial constraints such as limited free space and object collisions. In this work, we propose APIVOT, a VLM-based planner that adaptively interleaves language and visual thoughts for long-horizon planning. APIVOT learns to leverage language for semantic reasoning, while using visual thoughts as imagined future states for internal verification of geometric feasibility. On long-horizon kitchen tasks, APIVOT outperforms general-purpose VLMs and prior planning frameworks, achieving the largest gains in spatially constrained settings. We find that APIVOT learns meaningful modality selection behavior, demonstrating that adaptive interleaving of vision-language thoughts improves both planning success and reasoning efficiency.

Problem

• long-horizon planning
• semantic reasoning
• geometric feasibility
• spatial constraints
• robot planning

Innovation

• adaptive interleaving
• vision-language planning
• geometric feasibility
• long-horizon robot planning
• visual thoughts