Learning Fast Adaptation with Meta Strategy Optimization
This addresses the challenge of real-world deployment for legged robots by improving fast adaptation, though it appears incremental as it builds on existing meta-learning and adaptation techniques.
The paper tackles the problem of enabling legged robots to quickly adapt to new walking scenarios, such as sim-to-real transfer or motor failures, by introducing Meta Strategy Optimization, which allows adaptation with a handful of trials and outperforms previous methods in real and simulated experiments.
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 meta-learning 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.