Error-Aware Policy Learning: Zero-Shot Generalization in Partially Observable Dynamic Environments
This addresses the challenge of deploying robotic systems in partially observable dynamic environments, offering a novel solution for zero-shot generalization, though it appears incremental in its application to existing sim-to-real frameworks.
The paper tackles the sim-to-real transfer problem in robotics by developing error-aware policies that adapt to new environments in a zero-shot manner, demonstrating successful transfer to unseen human agents with different biomechanical characteristics and other control tasks.
Simulation provides a safe and efficient way to generate useful data for learning complex robotic tasks. However, matching simulation and real-world dynamics can be quite challenging, especially for systems that have a large number of unobserved or unmeasurable parameters, which may lie in the robot dynamics itself or in the environment with which the robot interacts. We introduce a novel approach to tackle such a sim-to-real problem by developing policies capable of adapting to new environments, in a zero-shot manner. Key to our approach is an error-aware policy (EAP) that is explicitly made aware of the effect of unobservable factors during training. An EAP takes as input the predicted future state error in the target environment, which is provided by an error-prediction function, simultaneously trained with the EAP. We validate our approach on an assistive walking device trained to help the human user recover from external pushes. We show that a trained EAP for a hip-torque assistive device can be transferred to different human agents with unseen biomechanical characteristics. In addition, we show that our method can be applied to other standard RL control tasks.