Event
CovNav Uses Conformal Prediction to Invoke Slow Reasoning on Demand for ObjectNav
1 reports · 1 sources
Pengying Wu et al. propose CovNav, a training-free framework that uses conformal prediction to turn spatial commonsense queries in ObjectNav into binary decisions: a local large model, in fast mode, judges whether the target is likely near already-observed objects, and only ambiguous queries are escalated to slow chain-of-thought reasoning. On Gibson and HM3D, CovNav's success rate is 0.8 and 1.7 percentage points higher than the strongest zero-shot baseline, while invoking slow reasoning for only 12.4% of queries, retaining most of the gains of always-on slow reasoning with substantially lower latency; real-robot experiments on a TurtleBot 4 also validated the method. Its object-pair formulation keeps calibration feasible and can support open-set objects via off-the-shelf detectors.
Report timeline
- CovNav: Conformal Prediction-Guided Adaptive Inference for ObjectNav
CovNav is a training-free framework that uses conformal prediction (CP) to turn spatial queries in ObjectNav into structured binary decisions: a local large model in fast mode judges whether the target is likely near an observed object, CP converts its token probabilities into a prediction set, and only ambiguous queries are escalated to slow chain-of-thought reasoning.
arXiv cs.RO ·