Behavior-aware merging improves continual agent learning success rates
BACAM: Behavior-Aware Continual Agent Merging for Multi-Turn Interaction
Artificial Intelligence
Summary
When AI agents learn new skills one after another, they often forget what they already know. The authors look at how to combine these specialized agents over time without needing to keep old ones around. They created BACAM, which uses feedback from how agents behave in interactions to carefully merge new skills while keeping old ones intact. This approach worked better than previous methods on tasks like shopping online, using tools, finding information, and moving through environments.
What this means in practice
- •For conversational ai developers: Integrate new dialogue skills into existing AI assistants sequentially without losing previously learned conversational behaviors.
- •For robotics engineers: Update robots with new interactive capabilities over time while preserving earlier skills, supporting longer-term adaptive behavior.
- •For customer support platform teams: Improve virtual agents by sequentially adding specialized task expertise without needing to retrain or store all prior models.$Commercial implications: Enables sale of adaptable customer service AI that incrementally expands capabilities without complete retraining.
Authors
Shuaitong Li, Baochen Xiong, Xiaoshan Yang, Xizhe Zheng, Yifan Xu, Jianhao Huang, Changsheng Xu
Abstract
Model merging offers a way to integrate the capabilities of specialized experts, but existing agent merging methods typically require all of them to be available at once. We study continual agent merging, which integrates incoming experts sequentially without retaining previously merged experts. Yet merging in parameter space or feature subspaces does not ensure that the merged model acquires an incoming expert's behavior on interaction trajectories. Moreover, updates toward a new expert can disrupt the merged model's previously integrated interactive behavior. Therefore, we propose Behavior-Aware Continual Agent Merging (BACAM), which learns parameter-wise merging gates from candidate-generated trajectories using expert-guided behavioral supervision. Task-level stability-plasticity control and tensor-level conflict-aware update budgets limit interference with existing capabilities while allowing new ones to be acquired. The learned gates are folded into the model weights without additional inference-time parameters. Across four interactive tasks - web shopping, tool use, information retrieval, and embodied interaction - BACAM achieves an average success rate of 62.82%, exceeding the strongest evaluated merging baseline by 21.69 percentage points. Our code is publicly available at https://github.com/shuaitongli/BACAM.