Chi, H., Feng, Z., LYU, Y., Zheng, C., Luo, L., Ong, Y. S., Tsang, I., Chen, H., Chang, Y., & Yin, H. (2026). InstructFlow: Adaptive Symbolic Constraint-Guided Code Generation for Long-Horizon Planning. Advances in Neural Information Processing Systems.
Abstract:
Long-horizon planning in robotic manipulation requires translating under-specified, symbolic goals into executable control programs that satisfy spatial, temporal, and physical constraints. However, existing language model-based planners often struggle with decomposing long-horizon tasks, enforcing constraints robustly, and adapting effectively to execution failures. We introduce InstructFlow, a multi-agent framework that establishes a symbolic, feedback-driven flow for code generation in robotic manipulation. InstructFlow comprises three coordinated agents: a InstructFlow Planner that constructs and traverses a hierarchical instruction graph to decompose goals into semantically grounded subtasks; a Code Generator that synthesizes executable code snippets conditioned on this graph; and a Constraint Generator that analyzes execution feedback to induce symbolic constraints when execution failures occur. These constraints are propagated upstream to refine the instruction graph and guide localized code revision without full regeneration. This graph-guided, dynamic flow enables structured, interpretable, and failure-resilient planning, yielding substantial improvements in task success rate and robustness across diverse manipulation benchmarks, particularly in constraint-sensitive and long-horizon scenarios. The implementation is available
at https://github.com/chiht21/InstructFlow.
License type:
Publisher Copyright
Funding Info:
This research / project is supported by the Agency for Science, Technology and Research (A*STAR) - Career Development Fund (CDF)
Grant Reference no. : C233312007
This research / project is supported by the Agency for Science, Technology and Research (A*STAR) - Career Development Fund (CDF)
Grant Reference no. : C243512014
This research / project is supported by the National Research Foundation, Singapore - AI Singapore Programme
Grant Reference no. : AISG-NMLP-2024-003