Computer Vision News - December 2021

7 Rapid Exploration for Open-World Navigation Results The authors experiment RECON on a mobile ground robot in open-world exploration scenarios. The results are quite exceptional given that they usually outperform state of the art methods in both time and robustness. They compare RECON to five baselines, trained on the same 20 hours of offline data, and finetuned in the target environment. RECON outperforms all baselines both in time spent discovering the desired goal and in time spent navigating to the discovered goal using prior exploration. As can be observed below, RECON achieves to navigate to the checkerboard goal, and uses a shorter path taking 30% less time than the only other successful method.