SELP is a novel framework that enhances large language models' capability to generate safe, effective task plans for robots while adhering to user-specified constraints.

• Combines equivalence voting, constrained decoding, and domain-specific fine-tuning to improve plan generation
• Ensures robots follow safety constraints even for complex, long-horizon tasks
• Maintains high performance across drone navigation and robotic manipulation scenarios
• Significantly outperforms existing methods in constraint adherence while maintaining efficiency

Engineering Impact: Enables more reliable deployment of LLM-powered robots in real-world settings where safety constraints and operational boundaries are critical to mission success.

SELP: Generating Safe and Efficient Task Plans for Robot Agents with Large Language Models