Learnable Conformal Prediction with Context-Aware Nonconformity Functions for Robotic Planning and Perception

Divake Kumar1, Sina Tayebati1, Francesco Migliarba1, Ranganath Krishnan2, Amit Ranjan Trivedi1
1University of Illinois at Chicago, 2Intel Labs
arXiv Demo Video Path Planning Code Detection Code Classification Code

Learnable Conformal Prediction provides context-aware uncertainty quantification for safe and efficient robotics, demonstrated on Intel NUC edge hardware with real-time performance.

Abstract

Deep learning models in robotics often output point estimates with poorly calibrated confidences, offering no native mechanism to quantify predictive reliability under novel, noisy, or out-of-distribution inputs. Conformal prediction (CP) addresses this gap by providing distribution-free coverage guarantees, yet its reliance on fixed nonconformity scores ignores context and can yield intervals that are overly conservative or unsafe.

We address this with Learnable Conformal Prediction (LCP), which replaces fixed scores with a lightweight neural function sθ(x) = fθ(φ(x)) that leverages geometric, semantic, and model cues. Trained to balance coverage, efficiency, and calibration, LCP preserves CP's finite-sample guarantees while producing intervals that adapt to instance difficulty, achieving context-aware uncertainty without ensembles or repeated inference.

On the MRPB benchmark, LCP raises navigation success to 91.5% versus 87.8% for Standard CP, while limiting path inflation to 4.5% compared to 12.2%. For object detection on COCO, BDD100K, and Cityscapes, it reduces mean interval width by 46–54% at 90% coverage, and on classification tasks (CIFAR-100, HAM10000, ImageNet) it shrinks prediction sets by 4.7–9.9%. The method is also computationally efficient, achieving real-time performance on resource-constrained edge hardware (Intel NUC with footprint 4.6 × 4.4 inch² and power <30W) while simultaneously providing uncertainty estimates along with the mean prediction.

Path Planning Visualization

Path Planning Visualization

Figure 1: Real-time conformal prediction for safe and efficient robotics. Left: Framework on Intel NUC mounted on Agilex MiniScout. Center: Conformal path planning with RRT. Right: Improved object detection and classification.

Path Planning Comparison

Office02 Environment

Shopping Mall Environment

Naive Office02

Naive Path Planning

Naive Shopping Mall

Naive Path Planning

Standard CP Office02

Standard CP (Fixed Margins)

Standard CP Shopping Mall

Standard CP (Fixed Margins)

Learnable CP Office02

Learnable CP (Adaptive Margins)

Learnable CP Shopping Mall

Learnable CP (Adaptive Margins)

Figure 2: Path planning benchmarks on MRPB. Rows: Naive, Standard CP, Learnable CP. Columns: Office02, Shopping Mall environments. Learnable CP adapts margins based on context.

Quantitative Results

Table I: Path Planning Results on MRPB

Path Planning Results Table

Table II: Object Detection Uncertainty Quantification

Object Detection Results Table

Object Detection with Uncertainty Bounds

Bounding Box Comparisons

Figure 3: Conformal prediction intervals across methods and datasets. Rows: COCO, BDD100K, Cityscapes. Columns: Standard, Ensemble, CQR, and Learnable CP. Red boxes show ground truth, green boxes show predictions with uncertainty bounds. Our Learnable CP provides tighter, more adaptive intervals.

Key Performance Metrics

  • Path Planning: 91.5% navigation success rate with only 4.5% path inflation (vs. 87.8% success and 12.2% inflation for Standard CP)
  • Object Detection: 46-54% reduction in mean interval width while maintaining 90% coverage on COCO, BDD100K, and Cityscapes
  • Classification: 4.7-9.9% reduction in prediction set sizes on CIFAR-100, HAM10000, and ImageNet
  • Real-time Performance: Achieves 39 FPS on Intel NUC edge hardware with <1% memory overhead

Demo Video

Real-time demonstration of Learnable Conformal Prediction on Intel NUC edge hardware with Agilex MiniScout robot

BibTeX

@misc{kumar2025learnableconformalpredictioncontextaware,
  title     = {Learnable Conformal Prediction with Context-Aware Nonconformity Functions for Robotic Planning and Perception}, 
  author    = {Divake Kumar and Sina Tayebati and Francesco Migliarba and Ranganath Krishnan and Amit Ranjan Trivedi},
  year      = {2025},
  eprint    = {2509.21955},
  archivePrefix = {arXiv},
  primaryClass = {cs.RO},
  url       = {https://arxiv.org/abs/2509.21955}
}