Search papers, labs, and topics across Lattice.
This paper introduces APIVOT, a vision-language model (VLM)-based planner that enhances long-horizon robot planning by adaptively interleaving language and visual thoughts. By leveraging language for semantic reasoning and visual thoughts for geometric feasibility verification, APIVOT significantly outperforms existing VLMs and planning frameworks, particularly in spatially constrained environments. The results indicate that this adaptive interleaving not only improves planning success rates but also enhances reasoning efficiency, showcasing a novel approach to complex robotic tasks.
Adaptive interleaving of vision and language thoughts in robot planning leads to significant improvements in task execution and reasoning efficiency.
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.
