Search papers, labs, and topics across Lattice.
This paper addresses the unreliability of Large Reasoning Language Models (LRLMs) in robotic task planning due to issues like overwriting valid plans and unresolved constraint violations during inference. The authors introduce SafeInferCom, a verifier-guided framework that monitors and intervenes mid-generation to preserve valid intermediate plans and correct errors without altering the LRLM's decoding trajectory. Experiments demonstrate SafeInferCom's ability to significantly improve planning success and accelerate error correction compared to one-shot inference, with further gains when combined with iterative refinement.
Mid-generation intervention during LRLM inference can preserve valid robotic plans and correct errors, boosting success rates and efficiency beyond one-shot or iterative refinement alone.
Large Reasoning Language Models (LRLMs) enable multi-step reasoning for robotic task planning, but continued reasoning can overwrite valid intermediate plans or leave constraint violations unresolved, reducing planning reliability and wasting inference-time computation. We develop an inference-time monitor that exposes and verifies intermediate plans without disrupting the original decoding trajectory. Building on this monitor, we propose SafeInferCom, a formal verifier-guided framework that preserves valid intermediate plans and directs error correction during generation. Experiments across multiple LRLMs and planning domains reveal reasoning-response inconsistency and limited self-correction under one-shot inference. SafeInferCom improves planning success and accelerates error correction relative to one-shot inference. When combined with iterative refinement, it further improves success while reducing token usage compared with refinement alone. We additionally evaluate SafeInferCom in VirtualHome and provide a real-world robotic-arm demonstration.
