CurricuLLM leverages large language models to automatically design progressive training curricula for robots, eliminating the need for human experts to manually design learning sequences.

• Creates structured learning paths that gradually increase task difficulty
• Enables robots to master complex skills that would be difficult to learn directly
• Demonstrates success in both simulated environments and real-world validation
• Reduces human engineering time while improving robot performance

This approach represents a major advancement for robotics engineering by making sophisticated skill acquisition more accessible and scalable across diverse applications—from manufacturing to healthcare.

CurricuLLM: Automatic Task Curricula Design for Learning Complex Robot Skills using Large Language Models