{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/learning-a-representation-map-for-robot","title":"Learning a Representation Map for Robot Navigation using Deep Variational Autoencoder","arxiv_id":"1807.02401","date":"2018-07-05","proceeding":null,"authors":["Kaixin Hu","Peter O'Connor"],"abstract":"The aim of this work is to use Variational Autoencoder (VAE) to learn a\nrepresentation of an indoor environment that can be used for robot navigation.\nWe use images extracted from a video, in which a camera takes a tour around a\nhouse, for training the VAE model with a 4 dimensional latent space. After the\nmodel is trained, each real frame has a corresponding representation point on\nmanifold in the latent space, and each representation point has corresponding\nreconstructed image. For the navigation problem, we map the starting image and\ndestination image to the latent space, then optimize a path on the learned\nmanifold connecting the two points, and finally map the path back through\ndecoder to a sequence of images. The ideal sequence of images should correspond\nto a route that is spatially continuous - i.e. neighbor images in the route\nshould correspond to neighbor locations in physical space. Such a route could\nbe used for navigation with computer vision techniques, i.e. a robot could\nfollow the image sequence from starting location to destination in the\nenvironment step by step. We implement this algorithm, but find in our\nexperimental results that the resulting route is not satisfactory. The route\nconsist of several discontinuous image frames along the ideal routes, so that\nthe route could not be followed by a robot with computer vision techniques in\npractice. In our evaluation, we propose two reasons for our failure to\nautomatically find continuous routes: (1) The VAE tends to capture global\nstructures, but discard the details; (2) the Euclidean similarity metric used\nfor measuring continuity between house images is sub-optimal. For further work,\nwe propose: trying other generative models like VAE-GANs which may be better at\nreconstructing the details to learn the representation map, and adjusting the\nsimilarity metric in the path selecting algorithm.","url_abs":"http://arxiv.org/abs/1807.02401v2","url_pdf":"http://arxiv.org/pdf/1807.02401v2.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"learning-a-representation-map-for-robot","repo_url":"https://github.com/augustkx/VAE_learning-a-representation-for-navigation","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"none","reach":null}],"tasks":[{"task_slug":"robot-navigation","task_name":"Robot Navigation"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}