Multi-Turn RL Performance of 3B, 7B, and 13B VLA Models on ALFWorld with Increasing Instruction Complexity

This report synthesises findings from 14 peer-reviewed papers addressing the following research question: How does the task success rate of 3B parameter VLA models trained with multi-turn RL compare to 7B and 13B counterparts on the ALFWorld benchmark when instruction complexity increases. We introduce SwiftSage, a novel agent framework inspired by the dual-process theory of human cognition, designed to excel in action planning for complex interactive reasoning tasks. SwiftSage integrates the strengths of behavior cloning and prompting large language models (LLMs). 13 claims were extracted from source literature; 12 were independently verified against retrieved documents. An automated multi-reviewer quality assessment produced a score of 8.7/10. This report is a machine-generated literature synthesis and does not constitute original research.

Research goal: How does the task success rate of 3B parameter VLA models trained with multi-turn RL compare to 7B and 13B counterparts on the ALFWorld benchmark when instruction complexity increases?

Autonomous literature synthesis. Automated review score: 8.7/10. Full text and citation available at Assignee Research.