Abstract

To facilitate reliable deployments of autonomous robots in the real world, Out-of-Distribution (OOD) detection capabilities are often required. A powerful approach for OOD detection is based on density estimation with Normalizing Flows (NFs). However, we find that prior work with NFs attempts to match the complex target distribution topologically with na ̈ıve base distributions leading to adverse implications. In this work, we circumvent this topological mismatch using an expressive class-conditional base distribution trained with an information-theoretic objective to match the required topology. The proposed method enjoys the merits of wide compatibility with existing learned models without any performance degradation and minimum computation overhead while enhancing OOD detection capabilities. We demonstrate superior results in density estimation and 2D object detection benchmarks in comparison with extensive baselines. Moreover, we showcase the applicability of the method with a real-robot deployment.

Results on Synthetic Data

• Qualitative comparison synthetic on density estimation for base distributions.

• Quantitative comparison synthetic on density estimation for different flow architectures w.r.t. KLD, i.e.,DKL(p(u, y)||pφ,ψ(u, y)).

Benchmarking on OOD in Object Detection

Real-Robot Experiment

The t-SNE visualization for (a) feature embeddings from the object detector (b) latents of the proposed learned base distribution cRSB and (c) the uni-modal Gaussian on the training set of Pascal-VOC-OS.

Bibtex:

@inproceedings{

feng2023topologymatching,

title={Topology-Matching Normalizing Flows for Out-of-Distribution Detection in Robot Learning},

author={Jianxiang Feng and Jongseok Lee and Simon Geisler and Stephan G{\"u}nnemann and Rudolph Triebel},

booktitle={7th Annual Conference on Robot Learning},

year={2023},

url={https://openreview.net/forum?id=BzjLaVvr955}

}