Robust Tool Use via Fission-GRPO: Learning to Recover from Execution Errors
This addresses the problem of unreliable real-world deployment of LLMs in multi-turn tool interactions, where execution errors are inevitable, by improving error recovery capabilities.
The paper tackles the brittleness of large language models in recovering from tool execution errors by proposing Fission-GRPO, a framework that converts errors into corrective supervision during reinforcement learning, resulting in a 5.7% absolute improvement in error recovery rate and a 4% overall accuracy gain for Qwen3-8B on the BFCL v4 benchmark.
Large language models (LLMs) can call tools effectively, yet they remain brittle in multi-turn execution: following a tool call error, smaller models often degenerate into repetitive invalid re-invocations, failing to interpret error feedback and self-correct. This brittleness hinders reliable real-world deployment, where the execution errors are inherently inevitable during tool interaction procedures. We identify a key limitation of current approaches: standard reinforcement learning (RL) treats errors as sparse negative rewards, providing no guidance on how to recover, while pre-collected synthetic error-correction datasets suffer from distribution mismatch with the model's on-policy error modes. To bridge this gap, we propose Fission-GRPO, a framework that converts execution errors into corrective supervision within the RL training loop. Our core mechanism fissions each failed trajectory into a new training instance by augmenting it with diagnostic feedback from a finetuned Error Simulator, then resampling recovery rollouts on-policy. This enables the model to learn from the precise errors it makes during exploration, rather than from static, pre-collected error cases. On the BFCL v4 Multi-Turn, Fission-GRPO improves the error recovery rate of Qwen3-8B by 5.7% absolute, crucially, yielding a 4% overall accuracy gain (42.75% to 46.75%) over GRPO and outperforming specialized tool-use agents.