ASPIRE: Agentic /Skills Discovery for Robotics

2026-06-30

AI Summary

Traditional robot programming struggles to integrate multimodal perception, contact dynamics, and diverse tasks while lacking autonomous repair capabilities. This work proposes ASPIRE, a system grounded in the “code-as-policy” paradigm that autonomously writes, diagnoses, and optimizes control programs through a continual learning loop, distilling effective repairs into reusable skills. ASPIRE establishes the first open-ended, agent-driven framework for skill discovery, enabling cross-task, cross-modal, and cross-embodiment skill transfer and demonstrating zero-shot generalization to unseen long-horizon tasks. Evaluated on LIBERO-Pro, RoboSuite, and BEHAVIOR-1K, ASPIRE improves success rates by 77%, 72%, and 32%, respectively. On LIBERO-Pro Long, it achieves a 31% zero-shot success rate—substantially outperforming the 4% baseline—and provides preliminary evidence of sim-to-real skill transfer.

Abstract

Traditional robot programming is challenging: it requires orchestrating multimodal perception, managing physical contact dynamics, and handling diverse configurations and execution failures. We introduce ASPIRE (Agentic Skill Programming through Iterative Robot Exploration), a continual learning system that autonomously writes and refines robot control programs in a code-as-policy paradigm while compounding experience into a reusable skill library. ASPIRE discovers skills that persist across tasks, simulation and real-world settings, and embodiments. It operates in an open-ended loop with three components: (1) a closed-loop robot execution engine that exposes fine-grained multimodal traces, enabling autonomous failure diagnosis, repair synthesis, and validation; (2) a continually expanding skill library that distills validated fixes into reusable, transferable knowledge; and (3) evolutionary search that generates diverse task sequences and control programs to explore beyond single-trajectory refinement. ASPIRE surpasses prior methods by up to 77% on LIBERO-Pro manipulation under perturbation, 72% on Robosuite bimanual handover, and 32% on BEHAVIOR-1K long-horizon household tasks. Its accumulated library also enables zero-shot generalization to unseen long-horizon tasks: on LIBERO-Pro Long, ASPIRE achieves 31% success versus 4% for prior methods despite their use of test-time reasoning and retries. Finally, simulation-discovered skills provide initial evidence of sim-to-real transfer, substantially reducing real-robot programming effort across different embodiments and robot APIs.

Problem

• robot programming
• multimodal perception
• contact dynamics
• execution failures
• skill generalization

Innovation

• code-as-policy
• continual learning
• skill library
• evolutionary search
• sim-to-real transfer