Autonomous Evaluation and Refinement of Digital Agents
This work addresses the challenge of evaluating and refining digital agents for tasks like web navigation and device control, offering a method to enhance performance without additional supervision.
The paper tackles the problem of improving digital agents' performance in web navigation and device control by using domain-general automatic evaluators, achieving a 29% improvement on WebArena and around 75% relative improvement in device control settings.
We show that domain-general automatic evaluators can significantly improve the performance of agents for web navigation and device control. We experiment with multiple evaluation models that trade off between inference cost, modularity of design, and accuracy. We validate the performance of these models in several popular benchmarks for digital agents, finding between 74.4 and 92.9% agreement with oracle evaluation metrics. Finally, we use these evaluators to improve the performance of existing agents via fine-tuning and inference-time guidance. Without any additional supervision, we improve state-of-the-art performance by 29% on the popular benchmark WebArena, and achieve around 75% relative improvement in device control settings.