Pigey

Pigey is a research project from Princeton Robotics (PhD student Liane Galanti, advisor Dhruv Shah) that combines LLM-based reasoning with existing robot policies as a zero-shot enhancer. It showed dramatic gains: real-robot performance from 16.7% to 97.3%, and simulation improvements on LIBERO-PRO from 12.8% to 53.3%. The approach requires no model retraining — LLM reasoning acts as a policy-level enhancer.