This is a page that curates AI-related papers published worldwide. All content here is summarized using Google Gemini and operated on a non-profit basis. Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.
From Tabula Rasa to Emergent Abilities: Discovering Robot Skills via Real-World Unsupervised Quality-Diversity
Created by
Haebom
Author
Luca Grillotti (AIRL, Imperial College London), Lisa Coiffard (AIRL, Imperial College London), Oscar Pang (AIRL, Imperial College London), Maxence Faldor (AIRL, Imperial College London), Antoine Cully (AIRL, Imperial College London)
Outline
This paper aims to achieve autonomous skill discovery, enabling robots to acquire diverse behaviors without explicit supervision. Existing Quality-Diversity Actor-Critic (QDAC) methods require manually defined skill spaces and carefully tuned heuristics, limiting their practical application. In this paper, we propose Unsupervised Real-World Skill Acquisition (URSA), an extension of QDAC, enabling robots to autonomously discover and master diverse and high-performance skills in real-world environments. We demonstrate that URSA successfully discovers diverse locomotion skills for the Unitree A1 quadruped robot in both simulated and real-world environments. It supports both heuristic-based skill discovery and fully unsupervised learning environments, demonstrating that the learned skill set can be reused for subsequent tasks such as real-world damage adaptation. We demonstrate that it outperforms baseline models in real-world damage scenarios, presenting a novel framework for real-world robot learning that enables continuous skill discovery with limited human intervention.
Takeaways, Limitations
•
Takeaways:
◦
A novel framework for autonomous skill discovery and mastery by robots in real-world environments.
◦
Addressing the need for manual skill space definition and heuristic tuning in QDAC's Limitations
◦
Supports both heuristic-based and fully unsupervised learning environments
◦
Proof of the reusability of learned skills in subsequent tasks, including adaptation to real-world damage.
◦
Contributes to improving the autonomy and adaptability of real-world robotic systems
•
Limitations:
◦
Only experimental results for the Unitree A1 quadruped robot are presented, and further research is needed to determine generalizability to other robot platforms.
◦
The scale and diversity of real-world experiments may be limited.
◦
Due to the nature of unsupervised learning, there is a possibility of unpredictability and stability issues in the learning process.
