Interactive Learning from Activity Description
This addresses the challenge of sample-efficient agent training for request-fulfilling tasks, offering a novel feedback mechanism that is more flexible than existing methods.
The paper tackles the problem of training request-fulfilling agents by introducing an interactive learning protocol that uses verbal activity descriptions for feedback, which improves sample efficiency compared to reinforcement learning and achieves competitive success rates without requiring demonstrations like imitation learning.
We present a novel interactive learning protocol that enables training request-fulfilling agents by verbally describing their activities. Unlike imitation learning (IL), our protocol allows the teaching agent to provide feedback in a language that is most appropriate for them. Compared with reward in reinforcement learning (RL), the description feedback is richer and allows for improved sample complexity. We develop a probabilistic framework and an algorithm that practically implements our protocol. Empirical results in two challenging request-fulfilling problems demonstrate the strengths of our approach: compared with RL baselines, it is more sample-efficient; compared with IL baselines, it achieves competitive success rates without requiring the teaching agent to be able to demonstrate the desired behavior using the learning agent's actions. Apart from empirical evaluation, we also provide theoretical guarantees for our algorithm under certain assumptions about the teacher and the environment.