Physics

Hill climbing

Archaeology

Key (lock)

Meta learning (computer science)

Mathematics

Structural engineering

History

Operating system

Optics

Machine learning

Milestone

Engineering

Mathematical analysis

Adaptation (eye)

Artificial intelligence

Systems engineering

Generalization

Variable (mathematics)

Programming language

Path (computing)

Computer science

Task (project management)

Process (computing)

Computer security

Robot

Climbing

IEEE Robotics and Automation Letters

The ability to walk in new scenarios is a key milestone on the path toward real-world applications of legged robots. In this work, we introduce Meta Strategy Optimization, a metalearning algorithm for training policies with latent variable inputs that can quickly adapt to new scenarios with a handful of trials in the target environment. The key idea behind MSO is to expose the same adaptation process, Strategy Optimization (SO), to both the training and testing phases. This allows MSO to effectively learn locomotion skills as well as a latent space that is suitable for fast adaptation. We evaluate our method on a real quadruped robot and demonstrate successful adaptation in various scenarios, including sim-to-real transfer, walking with a weakened motor, or climbing up a slope. Furthermore, we quantitatively analyze the generalization capability of the trained policy in simulated environments. Both real and simulated experiments show that our method outperforms previous methods in adaptation to novel tasks.

Learning Fast Adaptation With Meta Strategy Optimization