@@ -39,14 +39,19 @@ Traditional RL frameworks for LLM agents treat all tokens in a trajectory equall
VAGEN addresses these challenges by focusing optimization on the most critical decision-making tokens and creating a more nuanced reward structure across interaction turns.
## Experimental Results
Our experiments on visual Sokoban using a Qwen-VL 3B model show:
- TRICO significantly outperforms RICO in visual agentic tasks
- Both selective token masking and cross-turn credit assignment contribute to performance gains
- AICO (Action-centric Interaction Chain Optimization), which uses only selective token masking, outperforms TRICO on simple tasks
- TRICO demonstrates superior exploration capabilities on more complex problems
