Build on Priors: Vision--Language--Guided Neuro-Symbolic Imitation Learning for Data-Efficient Real-World Robot Manipulation

Researchers have developed a neuro-symbolic framework that enables robots to learn complex manipulation tasks from as few as one demonstration, without requiring manual programming or large datasets. The system uses Vision-Language Models to automatically construct symbolic planning domains and has been validated on real industrial equipment including forklifts and robotic arms.

• →The framework can teach robots complex tasks using only 1-30 demonstrations without manual domain engineering or semantic labeling.
• →Vision-Language Models are used to automatically classify skills and identify equivalent high-level states for autonomous learning.
• →The system was successfully tested on real industrial forklifts and Kinova Gen3 robotic arms across standard benchmarks.
• →Control policies are learned at the reference level rather than raw actuator signals, creating smoother and less noisy learning targets.
• →The approach enables data augmentation by projecting demonstrations onto different objects in the scene for enhanced learning.